What shipped, and when.

We took Austrian case law offline on 1 October while we removed criminal decisions from the general service. It is back.

Your agent can search the Rechtssätze, the courts' own condensed statements of the legal principle each decision settled, from the Oberster Gerichtshof, the Verwaltungsgerichtshof, the Verfassungsgerichtshof and the ordinary courts of appeal. The text comes from the Rechtsinformationssystem of the Bundeskanzleramt. We removed criminal, administrative-offence and criminal-derived decisions using the RIS's own metadata, and screened the rest for personal data before serving it.

Try it:

Using Ansvar, what has the Oberster Gerichtshof held on the duty to warn under Austrian contract law?

Austrian case law is available from Premium. Scope with sources=['austrian-court-decisions'], or let an Austrian search reach it.

Bulgarian searches reached the statutes only. A litigator asking how the Supreme Court of Cassation reads unjust enrichment or a contract term got the act and no court.

Your agent can now search the civil and commercial decisions of the Върховен касационен съд as the court publishes them on vks.bg, and fetch any decision back by its reference. Criminal decisions are excluded.

We read each decision for personal data before we serve it. The court names its judges, prosecutors and secretaries, and we keep them, as the court's own publication rules do. Where a decision names a private person, we remove that passage and leave the marker "[заличено]" in its place, so you can see that text was taken out. A decision we cannot clear that way we do not serve. verify_citations refuses a quotation that runs across a removed passage, so a quote never stitches the text on both sides together.

Try it:

Using Ansvar, how has the Bulgarian Supreme Court of Cassation ruled on неоснователно обогатяване between companies?

Bulgarian case law is available from Premium. Scope with sources=['bulgarian-court-decisions'].

Your agent could read EU legislation in English only. A Dutch lawyer quoting the AVG, or a German team checking the wording of the NIS2 Richtlinie, had to translate back from English or leave the gateway to find the official language version.

get_provision now accepts language for EU legislation, as one code or a list of up to three: en, nl, de or bg. A list returns the versions side by side under editions[]. Each version has its own rows, its own citation and its own language_served, so your agent can quote the German text and cite the German edition. Where an instrument has no edition in the language you asked for, the answer falls back to English and says so in language_fallback.

Search stays in English: search in English, then fetch the hits in the language you need. In verify_citations, each citation subject takes a language code, so a quote checks against the edition it came from.

Try it:

Using Ansvar, show me Article 17 of the GDPR in English and Dutch side by side.

describe_capabilities(section='sources') lists which editions each corpus holds.

A Luxembourg search reached the CSSF's circulars, but the circulars point at CSSF regulations and at statutes we served only as their amending acts. Your agent could read a circular on fund delegation and not the law it applies.

Your agent can now search and cite CSSF regulations, taken from the CSSF's own published files, and the CSSF's coordinated versions of these statutes in French and English:

• the Law of 5 April 1993 on the financial sector;
• the Law of 12 November 2004 on the fight against money laundering and terrorist financing;
• the Law of 17 December 2010 relating to undertakings for collective investment;
• the Law of 13 February 2007 relating to specialised investment funds;
• the Law of 12 July 2013 on alternative investment fund managers;
• the Law of 23 July 2016 on reserved alternative investment funds.

The CSSF drew up these consolidations for information. Every statute row says so: "CSSF coordinated text (unofficial); official text: Mémorial A / Legilux". Cite the Mémorial A for a filing; use these texts to find and read the rule.

Try it:

Using Ansvar, what does the Luxembourg Law of 17 December 2010 require of a UCITS management company on delegation?

Scope with jurisdictions=['LU'].

A Premium run of a threat model or a DPIA ended in a JSON report. The free trial run got a watermarked HTML or PDF. That put a paying Premium customer behind a free one, and we fixed it.

On Premium, generate_report now takes format html, pdf, docx or all and returns the rendered report without a watermark, the same report Team receives. Nothing else on Premium changes: document upload stays on Team and above, and the monthly run allowance is unchanged.

Ask your agent at the end of a run:

Using Ansvar, generate the report for this DPIA as a Word document.

Federally regulated banks and insurers in Canada work to these OSFI guidelines when they assess technology and supplier risk: B-13, Technology and Cyber Risk Management; B-10, Third-Party Risk Management; and E-21, Operational Risk Management and Resilience. Your agent can now search and cite them through the same financial regulation layer that serves the European regulators.

OSFI's guidelines are Crown copyright, so we asked before we fetched. OSFI granted written permission to reproduce the guidelines as published on the date of the grant. Every row carries OSFI's credit line, and we serve the text as OSFI published it. When OSFI issues a new edition of a guideline, we withdraw that guideline and ask again before we serve the new text.

Try it:

Using Ansvar, what does OSFI Guideline B-10 expect a bank to do before it outsources a critical service?

Scope a search to Canada with jurisdictions=['CA']. The guidelines are available on every plan.

Agents sometimes paste a whole case description into search: the parties, the facts, the dispute, a paragraph or more. The corpora behind search refuse a query longer than 500 characters. Until today that refusal came back from inside the corpus as an untyped error, and the gateway counted it as a source failure. The agent saw "all resolved MCP endpoints failed", or a case-law leg reported as unavailable, and had every reason to read it as an outage. Repeated refusals could also hold that corpus back from other searches for a short while.

The gateway now checks the length itself. A search or search_guidance query over 500 characters is refused before any source is searched, so no corpus is called and none is marked as failing. The tool result has isError: true and carries this body:

{
  "error": "validation_error",
  "code": "query_too_long",
  "max_chars": 500,
  "query_chars": 712,
  "message": "query is 712 characters; search accepts at most 500, so nothing was searched. Split the matter into separate issue queries of 2-5 legal terms, one call per issue, and merge the results. Do not send a truncated copy of the text: a cut-down description does not search better.",
  "dispatched": false,
  "automatic_retry": false
}

The query parameter of both tools now states the limit, so an agent that reads tools/list knows it before the first call. If your client caches tool definitions, refresh the connector to pick up the new wording.

What your agent should do with a case description: list the legal issues it raises and search each one in its own call, with 2 to 5 legal terms in the language of the corpus, then merge the answers. Do not send a shortened copy of the description. We tested that on Dutch case law: the opening of an intake text names the parties and the setting, the search ranks on those words, and the decisions that matter rarely come back. Short issue queries were the only approach we tested that found them in useful numbers.

Hand your agent the pattern with the case:

Using Ansvar, list the legal issues this case description raises and search Dutch case law for each one separately with a few legal terms, such as "ontslag op staande voet" or "dringende reden". Merge the results and cite each decision.

The limit and the error are on the tool reference.

The ansvar-compliance-skills plugin passed Anthropic's directory review and is now listed there. Until now you installed our agent skills one file at a time. Now one step adds the task skills listed below, together with the connector they need.

In claude.ai or the Claude desktop app, open Customize, then Plugins, find Ansvar Compliance Skills under Discover and select Add. The plugin is saved to your account, so Cowork loads it in your next task and Claude Code picks it up when you sign in with the same account. Directory plugins show on the Pro, Max, Team and Enterprise plans. In Claude Code you can also install it straight from its own repository, which doubles as a marketplace: /plugin marketplace add Ansvar-Systems/ansvar-compliance-skills, then /plugin install ansvar-compliance-skills@ansvar-compliance-skills.

The plugin bundles the Regulatory Threat Model, Incident Reporting Navigator, CRA Vulnerability Obligations and ISO Standards Expert skills, each vendored unchanged from its own published repository. A new release that also carries the workflow family skills is in preparation.

The skills do not answer from model memory. Each one calls the Ansvar gateway for the law it cites, so the plugin needs the connector signed in to your Ansvar account: open the plugin's Connectors tab and connect it. The incident and CRA skills run fully on a free account. The threat-model and ISO skills have plan notes, listed per skill in the guide.

Try it after connecting: "Using Ansvar, we have an incident. Who must we notify, in which member state, and by when?" Install steps for every client are in the agent skills guide.

The Slovenian statutes corpus grew from a small starter set into the core of the country's in-force legislation. You can now query the Obligations Code (OZ), the Companies Act (ZGD-1), the Employment Relationships Act (ZDR-1), the Public Procurement Act (ZJN-3), the income tax and VAT acts, civil, criminal, enforcement and administrative procedure, the insolvency act, the Personal Data Protection Act (ZVOP-2), the Information Security Act (ZInfV-1), consumer protection and electronic communications, alongside the Constitution. Each act comes from PISRS, the legal information system run by the Government Legislation Office, in its current consolidated version.

We read the licence first. Slovenian copyright law excludes official legislative texts from protection, and the Government Legislation Office publishes the consolidated collection on the national open-data portal under CC BY 4.0. Every row names the office as licensor, carries the licence notice and records that we split each act into provisions. We serve the Slovenian original only. We fetch act by act from a fixed, reviewed list, and we have asked the office for a key to its developer API.

Try it: "Using Ansvar, what does the Slovenian Employment Relationships Act say about the notice period for dismissal?" or in Slovenian: "Using Ansvar, kaj Zakon o gospodarskih družbah določa o odgovornosti članov uprave?" You can also look up an article directly, for example Article 131 of the Obligations Code, the general rule on liability for damage.

Consolidated texts in PISRS are unofficial; the Official Gazette (Uradni list) remains the authentic source, and each answer links back to the act's PISRS page. Two new judiciary acts, the Courts Act (ZS-1) and the Judges Act (ZSod), apply from 1 January 2027. We serve the acts they replace until then and add the new ones before that date.

An agent can cite a real article and still misquote it, or draw a firm conclusion from an assumption it never named. validate_claim catches both before the conclusion reaches your report.

Your agent sends its reasoning as a claim graph. A clause node carries the article it rests on and the quote taken from it. A fact node records something you told the agent about your organisation. An assumption node records something the agent assumed. Interpretations and conclusions point at the nodes they depend on and say whether the text settles them or whether they are a reading.

The gateway refetches every cited provision under your own tier and add-ons, so a source you cannot reach comes back as withheld, never as content. It compares each quote with the served text and marks the clause bound or not. Strength then flows through the graph: a conclusion carries the strength its weakest input allows, and a conclusion that rests on a fact or an assumption names that dependency. Every response lists what it checked and what it did not.

A quote in a different language from the served text counts as a reading at best, and a row served without language metadata stays unverified rather than assumed to match. Send the prose answer alongside the graph and the gateway compares its quotations too. Leave it out and the response says so under not_checked.

The response opens with a summary the gateway writes itself: one line per conclusion saying what it rests on (your input, an assumption, matched provisions, resolved evidence whose support is not assessed, or nothing this check could establish), a flag when a conclusion declared settled came back weaker, an attention list carrying every finding whether or not it changed a conclusion, and a relay text your agent is asked to pass on unchanged. The word "unverified" does not appear in it: the summary names the operation performed, never a verdict on the claim.

Every response also states what the check cannot do. It does not prove a row was served to you in an earlier call. It does not judge whether the logic translates the law correctly, since a wrong reading can bind to real text. It does not verify that a provision is in force today. Read the output as evidence for your review, not as a verdict.

The tool is available on every tier. A large graph citing one corpus takes tens of seconds. When the refetch budget runs out, the response is marked bounded, names how many clauses it did not reach, and leaves them unverified.

Try it once your agent has fetched the statute text: "Using Ansvar, build a claim graph for that conclusion and run validate_claim on it before you state it."

Ansvar has grown into more than a search over law, and the old site had stopped explaining it. The new one names four products and the one question each answers: what the law says (legal and regulatory evidence), what your system is (living security architecture), what you must do and what your risk is (compliance assessments), and how you run the management system (ISMS as Code). All four reach you through the same MCP connection, in the AI client you already use.

Deploy is now a page of its own, with hosted, your own Kubernetes, and on-prem or air-gapped side by side, a shared-responsibility table, and the pilot terms. Customer-managed deployment stays at the design-partner stage until the production scope is agreed, and the page says so.

Pricing keeps the plans you know and adds a table that shows which product each plan carries, so nothing has to be bought twice. Team is priced per workspace. Two offers are new and provisioned on request while we measure demand: Evidence Pro for one person who needs the interpretive layer, and an architecture workspace licence, self-hosted. Every quota, rate and ceiling the gateway enforces is on the limits page, with the exceptions written out.

Licensed ISO standards from SIS are named where they belong: an official distributor licence, per seat, on any plan. The proof we had let slip off the page is back: the OWASP GenAI landscape listing, the Regenold and W&B workflow, the CSA STAR self-assessment, the worked examples and the full sample deliverables.

The site carries far less text than before. Where a paragraph used to describe the product, a real artefact now shows it: a cited answer as Claude returns it, a data-flow model with a proposal awaiting assent, a gap finding with its evidence links, a pull request in an ISMS repository. Try the first one yourself: "Using Ansvar, what does the GDPR say about the right to erasure?"

The Canadian Centre for Cyber Security publishes the guidance Canadian organisations are asked to follow: the practitioner series (ITSP), the management series (ITSM), the awareness series (ITSAP), the ransomware playbook, and the threat bulletins and assessments. You can now query them through the gateway, in either official language, and every answer cites the Cyber Centre's page it came from.

We read the licence before we ingested a single page. The Cyber Centre's website terms permit non-commercial reproduction only, so on those terms alone the guidance stays out. The Government of Canada's Open Government Portal catalogues the same publications and declares each one under the Open Government Licence – Canada, which permits commercial use with attribution. We serve exactly the publications the portal catalogues, carry the portal record on every row, and have asked the Communications Security Establishment to confirm the reading in writing. A written objection withdraws the corpus.

Try it: "Using Ansvar, what does the Cyber Centre's ransomware playbook say about backups?" or in French: "Using Ansvar, que recommande le Centre pour la cybersécurité pour sauvegarder et récupérer les données?" You can also look a publication up by its number, for example ITSP.10.033, the security and privacy controls catalogue.

Two publications the portal lists point at pages the Cyber Centre has taken down, so they are absent until the publisher restores them. The Harmonized Threat and Risk Assessment methodology sits in the Cyber Centre's tools section rather than its guidance, and stays out for now. Canada's privacy regulators, federal and provincial, publish under terms that require written permission for commercial reuse; we have written to each of them, and their guidance joins when they answer.

You type Beurteilungen; the German section is headed Beurteilung der Arbeitsbedingungen. You type firmas electrónicas cualificadas; the Spanish article says firma electrónica ... cualificada. Those plurals missed the provision.

We now index every word in these corpora by its stem as well as by its spelling. A plural, a definite form or a case ending finds the provision that carries the base word. A query that matched before still returns the same rows.

We shipped this in waves over the past week, one corpus at a time: German, Austrian, Liechtenstein and German state (Landesrecht) legislation, then Dutch, Spanish, Italian, French, Portuguese, Luxembourg, Belgian, Czech, Polish, Finnish, Estonian, Romanian, Hungarian and Lithuanian law, together with the national drone-rules corpora for Germany, Spain, Belgium, Luxembourg, Hungary and Lithuania.

Swedish, Danish and Norwegian law joined them on 13 September.

The law itself did not move. We rebuilt the search index over the same text, so the provisions, their citations, the publisher links and the content dates are the ones served the day before.

Ask in the language of the corpus, in whichever form fits your sentence:

Using Ansvar, was sagen die deutschen Vorschriften über Beurteilungen der Arbeitsbedingungen?

Using Ansvar, ¿qué exige la ley española sobre firmas electrónicas cualificadas?

Using Ansvar, vad säger svensk lag om behandlingar av personuppgifterna?

Stemmers differ by language. German, Dutch, Spanish, Italian, French and Czech fold most of the inflection you would type. Finnish, Estonian, Polish, Hungarian, Romanian and Portuguese run on lighter stemmers: inflected forms are matched, and some endings still need the dictionary form. When a query in one of those languages comes back thin, try the base word.

The coverage page lists the jurisdictions behind these corpora.

The EIOPA lane added on 11 September served four document families. It now serves six: EIOPA statements and technical-standards reports (draft and amended implementing and regulatory technical standards) join the guidelines, opinions, supervisory statements and final reports. The lane also picks up EIOPA opinions the publisher's listing never typed, back to 2012, and the Joint ESA products EIOPA hosts, such as the supervisory statements on the Sustainable Finance Disclosure Regulation, the PRIIPs key information document and the DORA technical standards. A joint product that the EBA or ESMA publications already carry is served once. Letters, research papers, impact assessments, consumer-testing reports, consultation papers and stakeholder-group documents are not part of the set. Documents from before EIOPA's founding in 2011 and statements written for the UK withdrawal or the COVID-19 period are left out by design.

The standalone eiopa-insurance source is retired. Its documents are served by this lane with full text and a link to each publication page; the old source id keeps working as an alias, so a saved query that names it still resolves.

Using Ansvar, find EIOPA's implementing technical standards on the insurance product information document and any joint ESA supervisory statement on the PRIIPs key information document. Retrieve the relevant publications, distinguish technical standards from guidance, cite each source with its publication date, and name any gaps in coverage or access.

The coverage page shows the publications available on your topic; pricing lists plan access to guidance search. The financial services guide explains the research and assessment paths.

The European preparatory-works source now carries the current legislature of nine parliaments. New this release: the Spanish Congreso de los Diputados (Boletín Oficial de las Cortes Generales, government and parliamentary bills of the fifteenth legislature), the UK Parliament (bill publications of the current session), the Italian Senato della Repubblica (the Akoma Ntoso archive of the nineteenth legislature), the Czech Poslanecká sněmovna (parliamentary prints of the tenth period), the Slovak Národná rada (bills and explanatory reports of the ninth term) and the Lithuanian Seimas (the drafts registered this year). The Bundestag and Bundesrat, the Austrian Nationalrat and the Storting were already served and continue.

Each row cites the parliament's own publication with its licence: the Spanish and Czech material under their statutory exemptions, the UK bills under the Open Parliament Licence, the Italian archive under the Senate's Creative Commons terms, the Slovak and Lithuanian material under their statutory public-domain rules. Nothing is taken from a mirror.

Closed legislatures moved to a companion archive source: the German Wahlperioden before the current one, the Bundesrat history, the Italian thirteenth to eighteenth legislatures and the Austrian twenty-sixth and twenty-seventh Gesetzgebungsperioden. The live source stays small and refreshes with each parliament; the archive is rebuilt only when a term closes. A search for a historical Drucksache reaches the archive the same way it reached the live source before.

Ask your agent: "Using Ansvar, what did the Spanish government's bill say about the scope of the new customer-service obligations?" The answer cites the BOCG publication, to the document, and the lookup returns the same row.

Preparatory works are available from Premium. Scope with sources=['europe-preparatory-works'] for the current terms or sources=['europe-preparatory-works-archive'] for closed ones, or let a jurisdiction search reach them. The coverage page shows the current set.

Ansvar now serves EIOPA's final supervisory guidance — guidelines, opinions, supervisory statements and technical-standard final reports — in the same source as the EBA and ESMA publications added on 10 September. Each result names the publisher, the document type and a link to the publication on eiopa.europa.eu. Consultation papers, stakeholder-group documents and joint-ESA products are not part of this set; the joint products belong with the EBA publications.

Using Ansvar, find EIOPA's guidance on the supervision of reinsurance placed with third-country reinsurers. Retrieve the relevant publications, distinguish guidance from legislation, cite each source with its publication date, and name any gaps in coverage or access.

The coverage page shows the publications available on your topic; pricing lists plan access to guidance search. The financial services guide explains the research and assessment paths.

Ansvar adds assessment workflows for DORA ICT contracts, the DORA register of information, MiCA CASP readiness and AMLR readiness. Your agent checks the workflow types available to your account, confirms the evidence needed for the selected assessment, and returns findings with supporting sources and missing evidence identified. Unsupported findings stay unresolved.

DORA ICT contracts and register of information assessments are available from Team. MiCA CASP readiness and AMLR readiness assessments are available from Premium.

Using Ansvar, list the financial assessment workflows available to my account. Help me select the workflow for a DORA ICT contract review, confirm the evidence it needs, and leave unsupported findings unresolved. If the workflow is unavailable, report that and stop.

The financial services guide explains the research sources and assessment paths. Pricing lists plan access, and the agent docs explain how to run workflows through your connected agent.

Ansvar now serves financial supervisory publications from EBA, ESMA and the joint ESAs. You can retrieve the available guidance and Q&A material through your connected agent, with a reference to the publisher's source. The coverage page shows the publications available on your topic; pricing lists plan access to guidance search.

Using Ansvar, check the available ESA guidance sources for DORA ICT risk. Retrieve the relevant publications, distinguish guidance from legislation, cite each source, and name any gaps in coverage or access.

The financial services guide explains the research and assessment paths. The existing ESMA sanctions register rows have a separate publication scope; a register entry does not establish access to the full enforcement decision.

German case law on the gateway was federal: the Bundesgerichtshof, the Bundesverwaltungsgericht, the Bundesfinanzhof, the Bundesarbeitsgericht, the Bundessozialgericht and the Bundesverfassungsgericht, as Rechtsprechung im Internet publishes them. Most German litigation never reaches those benches. Ask about a dismissal dispute or a tenancy case and your agent found the federal court's last word on the question, or nothing.

The corpus now serves 281,813 decisions from two publishers. Beside the 84,156 federal decisions sit 197,657 from NRWE, the Rechtsprechungsdatenbank of Justiz NRW: the Oberlandesgerichte, Landgerichte and Amtsgerichte of North Rhine-Westphalia, and the state's administrative, labour, social and finance courts. Each row names the publisher that issued it, and your agent can fetch any decision it finds back from that publisher by its reference.

Ask your agent: "Using Ansvar, how have the North Rhine-Westphalia courts decided on Kündigungsschutz for employees on parental leave?" It answers from the Landesarbeitsgericht and the Arbeitsgerichte, cited to the decision.

Both publishers place their decisions outside copyright as amtliche Werke under UrhG §5, and both anonymise before publication. German court decisions are available from Premium. Scope with sources=['german-court-decisions'], or let a German or EU-wide search reach them. The coverage page shows the current set.

The gateway now serves the FCA Handbook: the rules, guidance and evidential provisions the Financial Conduct Authority publishes for UK financial services firms, from the Principles for Businesses and the Senior Management Arrangements, Systems and Controls sourcebook through the conduct, prudential and specialist sourcebooks. Each provision is served as the FCA's Handbook API publishes it, under the Open Government Licence v3.0, and every row cites the section on handbook.fca.org.uk.

The Handbook joins the European financial-regulation corpus, so a question about a UK firm's obligations can be answered beside the EU acts the same firm meets abroad, and a gap analysis or threat model for a regulated firm can cite the Handbook provision by its number.

Ask your agent: "Using Ansvar, what does SYSC 4.1 require of a firm's governance arrangements?" The answer comes from the Handbook, cited to the section.

The FCA's technical standards and the Handbook glossary are not yet served; they follow when the FCA's endpoints for them answer. The PRA Rulebook is a separate publication under its own licence and is not part of this release.

A NIS2 question about a member state used to come back as the directive. From today, a jurisdiction-scoped lookup of Article 20, 21 or 23 serves the national transposing provision where one is verified: the Cyberbeveiligingswet for the Netherlands, the loi NIS2 for Belgium, the BSI-Gesetz for Germany, the cybersäkerhetslagen for Sweden, the NISG 2026 for Austria and their counterparts across the Union, each cited to the article of the national gazette text.

The mapping is deliberately narrow. A whole article is served only where the national statute carries every paragraph the directive does; otherwise the paragraph that is verified is served and the response says so. The reporting stages of Article 23(4), early warning, notification, intermediate report, final report and progress report, resolve to the national provision for each stage where the statute splits them. Where a state has mapped only part of an article, the gateway serves the EU text and names the mapped paragraphs instead of claiming there is no mapping.

Ask your agent: "Using Ansvar, what does Belgian law require for the 72-hour incident notification under NIS2?" The answer is Article 35 of the loi du 26 avril 2024, cited to Justel, with the directive available by asking for the EU text.

Portugal joins the same day. The Regime Jurídico da Cibersegurança annexed to Decreto-Lei n.º 125/2025 is served whole again, article by article, after a parser repair, and its governance, risk-management and reporting articles are mapped like the rest.

National transposition serving is available on every tier. The coverage page shows the jurisdictions with a served statute; remaining gaps stay visible there rather than being papered over.

Chinese product teams preparing for European markets can now find Ansvar's introduction in Simplified Chinese. The page shows how to bring EU regulatory sources and Chinese national-law evidence into the same review, with each jurisdiction's requirements assessed separately.

The China coverage record lists the served national layer from the National Database of Laws and Regulations (flk.npc.gov.cn): laws of the National People's Congress and its Standing Committee, legislative decisions, and State Council administrative regulations. Text remains in Chinese, with Ansvar-authored English discovery metadata and citations to the original source. This is an unofficial reproduction; consult the promulgating organ's record. Where portal and corresponding paper text differ, the paper text prevails.

For example: "Using Ansvar, retrieve GDPR Article 32 and Article 51 of China's Personal Information Protection Law. Keep each source's original text, publisher and URL. Ask what you need to know about our processing activities before assessing applicability, and list the evidence we should gather for each requirement separately."

Local regulations, judicial interpretations and departmental rules are not yet served. GB/GB-T standard body text is excluded, and a reference to a Chinese standard does not establish equivalence with an ISO standard. Licensed ISO clause text requires the relevant entitlement.

The evaluation pack carries these national-layer limits too. Mainland connectivity, AI-client availability and working language are confirmed with each prospective customer; the Chinese introduction does not change those arrangements.

UK cyber security obligations are written in two registers. The statutes (the Network and Information Systems Regulations, the Computer Misuse Act, the Data Protection Act) set the duty; the National Cyber Security Centre explains how to meet it, in guidance collections that regulators, insurers and procurement teams treat as the reference. That second register is now searchable through the gateway beside the UK statutes already served.

Every page comes from ncsc.gov.uk itself, under the Open Government Licence v3.0, with the source URL, publisher and licence on each row. The crawl ran against the NCSC's own sitemap at a declared pace with an identifying user agent, and a page is only in the corpus when the full text was retrieved and retained; holding pages, navigation indexes and forms are excluded and recorded, never served as content.

Try it: "Using Ansvar, what does the NCSC say a board should ask about ransomware readiness, and which Cyber Assessment Framework outcomes does that map to?" Or scope a search to the United Kingdom with the cyber security sector and a plain term such as supply chain.

The UK cyber corpus sits in the same consolidated European cyber security service as the Finnish, Dutch and Swedish authorities, so a NIS2 gap analysis for a group with a UK subsidiary can cite the NCSC guidance beside the national regulators' text in one pass.

Swedish financial regulation lives on two levels. Riksdagen's acts (lagen om bank- och finansieringsrörelse, värdepappersmarknadslagen, betaltjänstlagen and the rest) set the frame and delegate the detail; Finansinspektionen's författningssamling (FFFS) fills it in — the föreskrifter a bank, fund manager or payment institution is examined against, and the allmänna råd that say how FI reads them. Until today the gateway served the acts only.

Every FFFS in force now serves from FI's published register: the consolidated text where FI publishes one, the amending instrument where it does not, each provision keyed by chapter and paragraph and cited to the exact PDF on fi.se with the retrieval date FI's terms ask for. Repealed instruments are left out rather than served as if still binding.

Try it: "Using Ansvar, what does FFFS 2014:1 require of a credit institution's internal control function, and which allmänna råd apply?" — or look a provision up directly, for example FFFS 2014:1 chapter 6.

The statutes and the rulebook answer together in one search, so a gap analysis or DPIA for a Swedish financial firm can cite the act, the FFFS provision under it, and FI's guidance in one pass.

The gateway now serves the Netherlands' wettelijk tuchtrecht: the rulings of the statutory tribunals that judge the regulated professions. Decisions of the medical colleges and the Centraal Tuchtcollege voor de Gezondheidszorg sit beside those of the tribunals for advocaten, notaries, accountants, gerechtsdeurwaarders, veterinarians and the maritime disciplinary court. All come from the KOOP Open Data Tuchtrecht collection, refreshed from the index the government publishes each day.

The gateway serves each ruling in full text, cited to its source and anonymised as the tribunal publishes it. These sit next to the Dutch statute law, court decisions and parliamentary history already on the gateway, so you can follow how a profession's own tribunal has reasoned, not only what the statute says.

Ask your agent: "Using Ansvar, how has the medical disciplinary tribunal ruled on informed consent when a patient later disputed the procedure?" The answer comes from the tuchtcollege's own decisions, cited case by case.

The rulings are available from Premium. Scope a search with sources=['dutch-tuchtrecht'], or let a Netherlands search reach them. The coverage page shows the current set.

Until this week the gateway served Dutch case law from the Hoge Raad alone. It now serves the decisions that rechtspraak.nl publishes as open data from every court in the roster: the highest courts, the gerechtshoven and the rechtbanken, with new decisions following the publisher's own releases.

search gained filters for case-law evidence. court takes the corpus's own court code (HR, RVS, CRVB, CBB, GHAMS, RBAMS and the rest), date_from and date_to bound the decision date as full ISO dates. Statute rows are unaffected; the filters are available from Premium, and a filter name the tool does not declare is refused with a structured error rather than ignored.

Ask your agent: "Using Ansvar, what has the Rechtbank Amsterdam decided on ontruimingsbescherming since January 2026?" The answer comes from the court's own decisions, each cited by ECLI with a deeplink to rechtspraak.nl.

Decisions are available from Premium. Scope a search with sources=['dutch-court-decisions'] or let a Netherlands search reach them, and fetch a decision by ECLI with get_decision. The coverage page shows the current set.

The gateway now serves the European Central Bank's legal acts in force: the regulations, decisions and guidelines the ECB adopts for the Eurosystem and for banking supervision, cited to the article as the Publications Office publishes them. The ECB's opinions on draft national and Union legislation sit beside them, back to the first years of the Bank, together with its memoranda and agreements.

Banking Supervision material joins in the same source. The ECB guides on capital and liquidity assessment, materiality and assessment methodology, the supervision newsletter articles, the public consultations, the supervisory sanctions register with each decision, the cyber resilience oversight expectations and the TIBER-EU framework are served from the ECB's own pages under the Bank's reuse terms, with the source notice on every row.

Ask your agent: "Using Ansvar, what does the ECB expect for DORA threat-led penetration testing?" The answer comes from the TIBER-EU guide and the ECB's own opinion on the regulation, cited to the section.

Legal acts and the sanctions register are on the free tier; the supervisory guidance is available from Premium. Scope with sources=['ecb'], or let an EU-wide search reach it. The coverage page shows the current set.

The Ansvar agent for Microsoft 365 Copilot passed Microsoft's store validation and is now listed in the Microsoft Teams and M365 Copilot store. Add it from the store (your tenant admin may need to allow it first), sign in with your Ansvar account, and the agent answers from the gateway with a citation to the official source on every answer. Copilot Studio and every other MCP client keep working as before: see Setup.

Italian national legislation now answers through the gateway, and Italy sits with the live jurisdictions on our coverage page. The corpus is the consolidated text published on Normattiva — the official legislation portal the Istituto Poligrafico e Zecca dello Stato runs for the Presidenza del Consiglio dei Ministri — and every row cites back to its Normattiva URN. Italy's data-protection shelf goes live alongside it: the Codice Privacy (D.Lgs. 196/2003 as amended by D.Lgs. 101/2018) and the Garante's provvedimento n. 467 of 11 October 2018, the list of processing that requires a DPIA under GDPR Article 35(4).

A jurisdiction reaches the coverage page only after we have read its licence ourselves, on the publisher's own pages. Italy rested on two findings, both re-verified against primary sources this week. Legge 633/1941 Article 5 places the texts of official acts of the Italian State and its public administrations outside copyright. And Normattiva's open-data terms moved to Creative Commons BY 4.0 on 1 January 2026, commercial use included — before that they were non-commercial. Normattiva's Avviso legale permits reproduction on three conditions our rows meet: name the source, and state that the text is not the authentic version and is available free of charge.

Try it: "Using Ansvar, which types of processing does the Garante's provvedimento 467/2018 make subject to a DPIA?" — or ask in Italian: "Using Ansvar, cosa prevede il Codice Privacy sul trattamento dei dati nel rapporto di lavoro?"

Court decisions are the known gap. Article 5 covers statutes and administrative acts; it does not reach judgments of the Corte di cassazione, the Consiglio di Stato or the Corte costituzionale, which need a separate basis. Italian case law stays out of the corpus until we have one.

The gateway now serves national consumer protection statutes for Austria, Germany, Estonia, Finland, Latvia, the Netherlands, Norway and Sweden: consumer sales and contract codes, unfair commercial practice acts, distance and off-premises contract rules, unfair contract terms, product safety law, and the consumer ADR and enforcement acts behind them. Every provision carries a citation to its national publisher.

The EU layer sits beside it. The Consumer Rights Directive, the Unfair Commercial Practices Directive, the Unfair Contract Terms Directive, the General Product Safety Regulation and the price indication rules are served at EU level, so your agent can read a directive and the national statute that transposes it in one conversation.

Ask your agent: "Using Ansvar, what does Swedish law require before a consumer can cancel a distance purchase?" The answer comes from the statutes as the Riksdag published them, cited to the provision.

Search in each statute's own language and scope by country. Coverage grows country by country; the coverage page shows the current set.

Ontario is Canada's largest provincial jurisdiction, and its law now answers through the gateway: the consolidated statutes and regulations published on e-Laws, served in both English and French, with every citation linking back to ontario.ca. The King's Printer for Ontario permits reproduction of statutes, regulations and judicial decisions without charge; our rows carry that attribution and the required not-an-official-version notice.

The same release refreshed the rest of the Canadian shelf. The federal corpus and British Columbia are current again, and the corpus data now records each Act's commencement status — groundwork for showing you, in a future release, when a provision you found is not yet in force. Canadian medical-device coverage (SOR/98-282) was refreshed to its 2026 amendments, and its sections now carry their published headings instead of bare section numbers.

Try it: "Using Ansvar, what does Ontario's PHIPA require before a hospital discloses personal health information?" — or ask in French: "Using Ansvar, que prévoit la Loi de 2010 sur la responsabilisation du secteur parapublic?"

Case law is the known gap: Supreme Court of Canada decisions are licence-cleared for reproduction but the publisher's access controls block automated retrieval, so we are pursuing the official route rather than working around it. Coverage lands when that conversation does.

Dutch public-sector bodies — Rijk, provincies, gemeenten, waterschappen — measure their information security against the Baseline Informatiebeveiliging Overheid 2 (BIO2). Version 1.3, as published in the Staatscourant, is now a corpus on the gateway: Deel 1 (the framework — obligations, ISMS, risk management, the statement of applicability, governance, suppliers, the Cbw relationship) and every Deel 2 overheidsmaatregel, served verbatim from the official publication.

Each measure carries its NEN-EN-ISO/IEC 27002:2022 control number and whether it falls under the Cyberbeveiligingsregeling sector overheid — the ministerial regulation, in force with the Cyberbeveiligingswet since 15 August 2026, that turns the BIO2 measures into the statutory minimum for the sector overheid. So a DPIA, gap analysis or threat model for a Dutch public body can cite the measure, the ISO control behind it, and the legal duty that binds it, in one pass.

Try it: "Using Ansvar, which BIO2 measures apply to logging and monitoring, and are they mandatory under the Cyberbeveiligingswet?" — or look a measure up directly by its number (for example 5.01.01).

The ISO 27002 controls themselves are not reproduced — the BIO2 references them by number, and so do we; the norm text stays with NEN.

The gateway passed OpenAI's app review and is now in the ChatGPT Apps directory (page renders logged-out; installing asks for a ChatGPT login). Add it from the directory, sign in with your Ansvar account, and call it with @Ansvar Gateway in any chat. Other clients: see Setup.

The gateway passed AWS Marketplace review and is publicly listed as Ansvar Gateway under Legal & Compliance and AI Security in the AI Agents & Tools catalog. The listing is free with no end date: subscribe, set up a free account at app.ansvar.eu, and connect any MCP client via OAuth — the same gateway, found through AWS procurement.

GDPR Article 35 is shared; the national lists that sit on top of it are not. A DPIA run now screens against the supervisory authority's own Article 35(4) material for 29 jurisdictions — the EU, Iceland, Liechtenstein, Norway, Switzerland and the United Kingdom — querying each in the authority's own language, because an English meta-query misses a list published in Greek, Lithuanian or Icelandic. What comes back keeps the status the authority gave it: Germany's published DSK list and Estonia's expressly indicative guidance are not the same thing, Cyprus's material is a pre-opinion advisory draft and is labelled as one, and Switzerland has no list mechanism at all — its screen runs on the statutory high-risk test in DSG Articles 22 and 23. Luxembourg, Malta and Slovakia are identified but not reproduced: the run records the source as unavailable rather than rebuilding it from model memory, and a negative screening outcome is never grounded on a list that was not actually fetched. The DPIA coverage register publishes the state of every jurisdiction — authority, source, legal status, verification date, known limitation. Try it: "Using Ansvar, run a GDPR Article 35 DPIA for our new HR analytics platform. Confirm the relevant jurisdictions with me, screen the applicable national trigger sources, then continue to the prior-consultation determination."

Every workflow family now has an installable agent skill: gap analysis, threat modelling, impact assessments (DPIA and FRIA), TARA, tender review, vulnerability decisions, adversary tabletop, document review, and SORA drone operations — nine in all. A skill is one markdown file the agent reads once and then applies in every conversation, and each of these carries the run loop (how to open a run, what each stage asks for, how to resume one), the delivery rules that govern what a finished report must contain and how the agent relays the receipt the server hands back, the family's own method, and starter questions to paste. They are not copy written for the website: each is generated from the same instruction library the workflow engine runs, then imported here by checksum — the published file, the release manifest that hashes it, and the pinned build are checked against each other in CI, so a skill cannot describe a workflow the engine no longer implements. Install from agent skills: take the SKILL.md or the ready-made ZIP, and connect the gateway. Try it: "Using Ansvar, run a DPIA for our new HR analytics platform and take it through to the prior-consultation determination."

"Using Ansvar, run a gap_analysis_mdr workflow for our device and assess the Annex I GSPRs against our current technical documentation." Your agent can now run that against a workflow we have proven in production, not just published. On 2026-08-11 each of the three medical-device workflows completed a full run on the live platform, from scoping interview to rendered report, and the runs were accepted on correctness: citations round-trip through the platform's lookup tools with their source, publisher and licence intact, and requirements without supporting evidence carry an explicit flag rather than a decorative citation.

gap_analysis_mdr assesses all General Safety and Performance Requirements of MDR Annex I, surfaces the Annex VIII device class because it sets the conformity route, and maps each requirement to the harmonised-standard clauses that demonstrate conformity: IEC 62304, IEC 81001-5-1, ISO 14971, IEC 62366-1 and their siblings. gap_analysis_ivdr does the same for IVDR Annex I, with analytical and clinical performance under Annex XIII treated as the load-bearing requirement. gap_analysis_mdcg_cyber assesses device cybersecurity against MDCG 2019-16 rev. 1 across secure design, technical documentation and post-market obligations, with MITRE ATT&CK available for adversary context.

The proof runs assessed fictional manufacturers built with deliberate weaknesses, and the point is what happened to those weaknesses: every planted gap surfaced as a finding, and the assessments found gaps beyond the planted ones, including one requirement the fixture had misclassified as inapplicable. Where the case-law layer returned nothing on point, the finding says case_law_unconfirmed: true instead of citing something adjacent. An assessment you would hand a notified body has to behave that way.

The guidance layer behind these workflows is current: the served corpus is reconciled against the Commission's MDCG guidance register including the 2026 series and revisions, and the MDR and IVDR harmonised-standards lists follow the June 2026 implementing decisions.

All three workflows are available from the Premium plan. Ask list_workflow_types for the catalog, or start with the prompt above.

"Using Ansvar, what regulatory deadlines hit us in the next 18 months?" Your agent can now answer that from a curated calendar rather than a client alert's prose. get_regulatory_deadlines returns dated obligation events, soonest first: application dates, commencement dates, transposition deadlines. Each row cites the provision that states the date, so you can read the source yourself.

The calendar tracks instruments as amended. Where an omnibus regulation moved a date, or a corrigendum corrected one, the row carries the corrected date and cites the document that states it. Rows name their date precision, record when a curator last verified them, and carry a day count computed at request time.

Ask for a jurisdiction and the answer includes the EU-level instruments that bind there. A query for Germany returns the AI Act's staged application dates alongside German national instruments, because an EU regulation binds in every member state. A request parameter turns that off.

Coverage at launch is curated and leans European. Every response names what was in scope, and states on every call that absence of a row is not absence of an obligation. The tool serves only from a signature-verified dataset: if a newer dataset fails verification, the previous one keeps serving and every response says so, and if none has ever loaded, the tool refuses with a typed error.

get_regulatory_deadlines is available from the Premium plan. Dataset state and freshness report through get_regulatory_intelligence_status, free on every plan.

Before you connect anything, your own AI can now tell you whether Ansvar is worth connecting. The new /evaluate page carries a single evaluation pack — copy it into Claude, ChatGPT, Copilot, Gemini, or any client your team already uses. No account, and no browsing capability needed: the pack is self-contained, with the audited coverage snapshot embedded.

Your agent asks four questions — jurisdictions, frameworks in play, which AI clients you run, what your agents should do — and then writes a fit report bound to the evidence in the pack. The verdict is one of four: documented fit, conditional fit, no current fit, or insufficient evidence. The rules are explicit and visible on the page: coverage claims may come only from the embedded snapshot, one missing non-negotiable jurisdiction outweighs any number of adjacent matches, and on a negative verdict the report must recommend against signing up rather than steer you to adjacent coverage. We wrote the pack, and it says so — vendor-authored evidence, weighed accordingly.

Agents that can fetch URLs find the same pack at /agent-evaluation.md, versioned copies alongside. On a fit, the report ends with the free verification path: connect on the Free plan and re-check the coverage it leaned on with live calls.

Your agent can now ask what has actually been published, and get an answer from records rather than from memory. Each one carries its title, a link to the original publisher page, the publication and observation dates, official identifiers where the source assigns them, and a full citation naming the licence and linking its terms.

What is monitored is the EU Official Journal L series — regulations, directives, implementing and delegated acts as they appear — alongside announcements from the European Commission, the EDPB and ENISA, and the national gazettes of a growing set of European jurisdictions. Every source passes the same per-source licensing gate before a single fetch, and get_regulatory_intelligence_status reports the live enrolment: which sources are watched, how fresh each one is, and when its baseline ran. That tool is the authoritative list, and it works on every plan including Free.

Coverage is bounded, and the tools say where the boundary is. Each source reports only from its own enrolment baseline onward, so an empty result means nothing matched inside that window — never that the regulators were quiet. When every selected source is stale or has never synced, the search fails and names the state of each one instead of returning an empty list that would read as silence.

This is publication monitoring, not amendment tracking: it tells you an act was published, not how a law you already follow changed. Provision-level amendment feeds are not available today, so the gateway withholds get_changes until a supported feed can answer it.

search_regulatory_updates and get_regulatory_update are available from the Premium plan. get_regulatory_intelligence_status is free on every plan.

The site now has a Workflows section covering both ways you can run the same work: yourself, in the AI client you already use, or as an engagement we run and review as practitioners. The first workflow family has its own page — TARA, the threat-analysis-and-risk-assessment family for vehicles, OT, rail, robots and drones, sharing one enterprise risk spine. Services keep their URL and now live in the Workflows menu.

The canonical control library also has a page for the first time. It explains what a canonical control is, why a flat framework tag cannot carry a coverage claim, the five relationships a mapping can record, and why the library stores citation pointers rather than licensed clause text. It is candid about where coverage stands: published crosswalks arrive untyped and claim nothing until a person reviews them, so the page shows what one control reaches today and names the frameworks that are registered but not yet mapped. A worked crosswalk from a real session walks ISO/IEC 27001 Annex A.5.26 through the spine to 24 NIST CSF 2.0 subcategories, with the provenance of every hop. The control library is available on Team and Company plans.

A service credential — the org-owned seat a Team or Company admin mints for headless clients such as n8n, CI, or a scheduled monitoring agent — now carries the same tool surface as a signed-in seat on its plan: search, workflow runs, document upload and report generation, under the same tier gates, drawing from the organization's pooled quota. This supersedes the read-only boundary the launch entry of 2026-07-26 described. Each credential keeps a monthly workflow-run ceiling, by default half of what one seat contributes to the pool, so one runaway agent cannot drain what the rest of the organization shares. A credential takes a seat of its own, next to yours; add seats as your agents grow. Setup guide: Service credentials.

Free-tier signup is now identity-native two ways: sign in with Microsoft Entra ID or Google as before, or register directly with an email address and password at app.ansvar.eu. Registration asks for terms acceptance up front, verifies the address by email before the account is usable, and sits behind a fail-closed bot check. Nothing about the tier changed — a free account still connects your own MCP client to the gateway with the same limits, and paid plans still upgrade through checkout. If your organisation blocks third-party SSO, this removes the last reason you couldn't try the connector.

The iso-standards-expert skill teaches an AI assistant to answer standards questions from served clause text instead of model memory: it fetches the clauses a question needs through the ISO Standards add-on, quotes them verbatim with the SIS copyright notice on every excerpt, and reports an unevaluated-scope list whenever the licence coverage ceiling withholds a fetch — only selected parts of a standard are ever displayed, never the complete standard. Without an add-on the skill still runs, answering ISO 27001-shaped questions from the free Secure Controls Framework cross-reference with every such result labelled. The sequence: install the skill, connect the Ansvar Gateway connector, sign in (a free account is enough), and subscribe per standard. Also available bundled in the ansvar-compliance-skills Claude Code plugin. Try it: "Using Ansvar, what does ISO/IEC 27001 require for a risk treatment plan, and what should ours contain?" Details in the agent skills guide.

A free account runs one workflow a month, picked from seven types: STRIDE threat model, generic gap analysis, NIS2, DORA, CRA and EU AI Act gap analyses, and DPIA. Solo doubles that to two runs. The report arrives as JSON or as a watermarked HTML or PDF render with a visible banner marking the inputs as self-asserted. Premium runs the full interview-grounded catalog; Team adds your own documents as evidence. Plans on Pricing.

Scope a search to Austria and you now get the Datenschutzbehörde's enforcement practice: 1,642 DSB and DSK decisions harvested from the official RIS registry, each linking its ris.bka.gv.at page. The corpus also adds 330 EDPB guidelines, recommendations and opinions; on Premium and higher plans these arrive inside EU-scoped search results and through the guidance tool. Both join the French CNIL deliberations already served, and every row cites the official publisher page.

A seat on Team and Company plans can now belong to an agent, not just a person. Org admins mint org-owned service credentials — OAuth2 client_credentials, read-only — for headless use: CI jobs, scheduled compliance checks, server-side agents. Each active credential takes a seat, the same way a person does. Manage them from the Service credentials section of Organization in the account area; plan details on Pricing.

Signing up for the free tier is now a plain sign-in: pick Google, Microsoft, or email at app.ansvar.eu/account and the account is provisioned on the spot. The €0 checkout flow is retired; existing free subscriptions keep working unchanged. Paid plans still go through Stripe checkout.

The gateway is now listed in Anthropic's connector directory (Claude login required to view). claude.ai and Claude Desktop users connect in one click; the listing also carries the Claude Code and Claude API snippets. Other clients: see Setup.

Ansvar Gateway is listed in the CSA STAR Registry: a Level 1 self-assessment, the full CAIQ v4.0.3, downloadable by anyone. We published all 261 answers, including the 67 No answers. Why we did it that way: the announcement post.