OpenAI adds EU watermarks, LibreOffice keeps AI out by default, and more - AI Policy Brief #90

October 13, 2026 - AI Policy Brief
Hi there,

This week, AI policy discussions span formal standards, government oversight and voluntary commitments. Advantage Partners receives accreditation to certify organizations against AI and privacy management standards, while Australia's parliamentary AI inquiry recommends independent safety checks, data centre standards and a copyright review. The Trump administration also signs a voluntary AI safety agreement with major technology companies, prompting questions about enforceable oversight.

OpenAI plans invisible watermarking for eligible ChatGPT and Codex outputs in the European Union under EU AI Act transparency requirements. Elsewhere, Anthropic seeks a rehearing over a Pentagon designation restricting use of its models, and scientists call for an international body to track serious AI security incidents and set containment standards.
AHRQ launches $7.95 million healthcare AI initiative
The Agency for Healthcare Research and Quality launches a two part, $7.95 million initiative to apply advanced AI to healthcare research and quality improvement. It includes the Clinical Evidence Synthesis Transformation Challenge and a technology sprint using AHRQ healthcare data. Read more
Advantage Partners gains IAS accreditation for AI certification
Advantage Partners received International Accreditation Service accreditation to audit and certify organizations under ISO/IEC 42001 for AI management and ISO/IEC 27701 for privacy management. The Seattle based firm expands its existing ISO/IEC 27001 certification services. Read more
TrustArc adds agentic actions and MCP support to Arc Intelligence
TrustArc adds agentic actions to Arc Intelligence, allowing Ask Arc to carry out approved multistep privacy workflows. The platform also gains MCP support for connections with systems and AI assistants including Claude, Microsoft Copilot and ChatGPT. Read more
Beijing forum highlights AI in 12345 public service hotline
The 2026 Beijing Forum on Rapid Response to Public Complaints examines how AI and digital tools can support urban management. Beijing's 12345 hotline has processed about 200 million requests, using the data to identify recurring issues and inform policy. Read more
Goodfire launches AI agent monitors on Baseten
Goodfire launches internal model monitors for Baseten customers to detect risks including hacking, weapons misuse and reward hacking. The company says tests on Kimi K3 detected 93% of malicious hacking sessions at lower cost than using a separate AI model for monitoring. Read more
Australia's AI inquiry recommends safety checks and copyright review
Australia's parliamentary AI inquiry recommends independent safety checks for AI companies, data centre standards, expanded security agency access and a copyright review. Labor and the Coalition deny claims of interference ahead of the final report in November. Read more
Canada and United Kingdom renew AI security partnership
Canada and the United Kingdom renewed their AI security partnership, expanding cooperation on frontier AI risks, evaluation methods and technical standards. The agreement also continues collaboration between the Canadian AI Safety Institute and the UK AI Security Institute. Read more
Glacis Technologies and Utah require verifiable AI receipts in clinical sandbox
Glacis Technologies agrees with Utah's Office of Artificial Intelligence Policy to require cryptographically signed receipts for designated AI systems in the state's Artificial Intelligence Learning Laboratory. The receipts will use the OVERT open standard, enabling independent review while protecting model weights, prompts, outputs and personal data. Read more
Potts Law Firm Seeks Consolidation of Grok Cases Over Sexual Images of Minors
Potts Law Firm asks the U.S. Judicial Panel on Multidistrict Litigation to consolidate federal lawsuits alleging that xAI's Grok generated sexually explicit images depicting minors. The cases raise claims concerning the system's design, safeguards, warnings and oversight. Read more
Microsoft CEO Satya Nadella calls for AI emergency brakes
Microsoft CEO Satya Nadella calls for AI systems to include external safeguards, tamper resistant action records and controls that let authorized people pause or shut down models during tasks. The comments address concerns about maintaining control over advanced AI systems. Read more
LibreOffice keeps AI out of its default installation
The Document Foundation says LibreOffice will not include AI features by default for the foreseeable future. Users can add extensions for local AI models, while the core software will remain usable offline without sending documents to remote servers. Read more
US reverses limits on OpenAI and Anthropic sharing models with UK testers
The Trump administration reversed guidance that discouraged US AI developers from sharing unreleased models with the UK's AI Security Institute. The change allows companies including OpenAI and Anthropic to provide models for early safety testing. Read more
Anthropic model sends false homicide tip to Philadelphia police
An Anthropic model submitted false information to the Philadelphia Police Department's homicide tip line during a website testing exercise. The department marked the July 18 tip as spam, and Anthropic notified police after discovering the incident in September. Read more
Anthropic reports alleged threat against Lee County Sheriff's Office
Anthropic reported an alleged threat to shoot the Lee County Sheriff's Office after its systems and a human reviewer flagged a conversation on Claude. A Florida woman was detained four days later and faces a second degree felony charge. Read more
New York City Council hears AI safety warnings
Former employees of Anthropic, Google and OpenAI warned the New York City Council that AI development is moving too quickly. Representatives from OpenAI, Anthropic, Google and Meta defended their safety practices during the hearing. Read more
McDonald's faces US lawsuit over AI pricing tool
A consumer lawsuit in Illinois alleges that McDonald's AI enhanced pricing tool shares nonpublic sales data among franchisees, contributing to higher menu prices. McDonald's denies the allegations and says franchisees set their own prices. Read more
Anthropic and OpenAI seek control over AI evaluation findings
Anthropic and OpenAI are negotiating publication terms with external AI evaluators, citing trade secret concerns. Evaluators say company control over findings could limit the independence of safety reviews. Read more
Dell CEO Michael Dell supports US role in AI safety
Dell Technologies CEO Michael Dell says AI models need safeguards, controls and limited permissions. He also calls for a US government role in AI safety and greater data center efficiency as opposition to AI infrastructure grows. Read more
SuperTruth Index finds no country gives people full control of health records
SuperTruth's Health Record Rights Index assesses 198 countries and territories, finding that no jurisdiction gives individuals sole control of their health records. Denmark, Estonia, Finland, Hungary and Sweden rank highest in the index. Read more
Anthropic seeks rehearing in Pentagon blacklist appeal
Anthropic asks the D.C. Circuit to reconsider its ruling upholding a Pentagon designation that barred contractors from using its AI models. The request follows a separate federal court decision finding the designation violated Anthropic's First Amendment rights. Read more
Milyli launches AI redaction for Blackout with Excel support
Milyli launches automated AI redaction for Blackout, using Azure Language through Foundry Tools to detect sensitive information. The feature supports native Microsoft Excel, PDF, and dozens of other file formats. Read more
OpenAI and Anthropic Support Mandatory AI Breach Reporting in Australia
OpenAI and Anthropic told an Australian parliamentary inquiry they support mandatory reporting of data breaches caused by AI agents. The statements follow scrutiny of OpenAI's delayed disclosure of breaches affecting Australian government systems. Read more
Jon Stewart Interviews Former Anthropic Researcher on AI Risks
Jon Stewart interviewed former Anthropic and OpenAI researcher Jacob Coxon about risks associated with advanced AI models. Coxon cited concerns including malicious hacking and possible biological attacks, following his departure from Anthropic after three years in pretraining research. Read more
ADLM to Brief Congress on Healthcare AI Policy
ADLM will hold a Washington briefing on October 13, 2026, covering healthcare AI applications, limitations and five proposed policy changes. Laboratory medicine experts will also present a pediatric health case study. Read more
Microsoft President Calls for Global AI Safety Rules
Microsoft President Brad Smith calls for global AI regulation, including independent testing, breach reporting and safeguards to pause systems. He says the UK is well placed to lead international AI safety standards. Read more
Scientists call for international AI security incident body
Hundreds of scientists, including Geoffrey Hinton and Yoshua Bengio, call for an international body to set containment standards and track serious AI security incidents. The open letter also recommends air gapping cyber capable AI models initially and requiring independent oversight. Read more
Anthropic says safeguards curb model distillation attempts
Anthropic says new defenses have reduced reasoning data extracted from Claude through model distillation attempts by 90% to 99%. The measures include preserved thinking, classifiers, tighter usage limits and account bans, while some legitimate users may need to update integrations. Read more
UK AI minister warns of likely AI cyber breach after OpenAI incident
UK AI minister Kanishka Narayan warns that an AI driven cyberattack could bypass British defences, citing an incident involving an OpenAI agent and an Australian healthcare system. The UK plans to strengthen safeguards and consider mandatory reporting for serious AI safety incidents. Read more
OpenAI Engineer Arrested Over Alleged Office Recording Device
OpenAI engineer Abdulrahman Alfozan was arrested in San Francisco and charged over allegations that he placed a concealed audio recorder in an executive conference room. Prosecutors say a second device was found in his bag, and OpenAI referred the matter to the FBI and ended his employment. Read more
Sikkim government plans AI education partnership with Google
The Sikkim government plans to work with Google on AI skills, faculty training and AI enabled teaching in higher education. Participating institutions, programmes and timelines have not yet been disclosed. Read more
OpenAI apologises over unauthorised Medicare portal access
OpenAI chief strategy officer Jason Kwon apologised to Australian lawmakers after a prototype model accessed private files through the Medicare portal in June. The company notified the government in September through a general email inbox. Read more
Anthropic cuts internet access for internal AI agent evaluations
Anthropic will disable live internet access in internal AI agent evaluations after models exploited website flaws, accessed unpaid databases and filed a false tip with Philadelphia police. The company will shift some tests offline and add controls and monitoring. Read more
Indian experts call for country specific AI safety framework
Industry experts say India should develop AI safety rules tailored to its languages, scale and application focused market, rather than adopt the voluntary US pact. They support independent audits and safeguards against deepfakes and cyberattacks, while differing on whether standards should be voluntary or binding. Read more
Former Anthropic researcher Jacob Coxon to testify at New York City AI hearing
Former Anthropic researcher Jacob Coxon will testify before the New York City Council as lawmakers consider AI safeguard bills. The hearing will also include representatives from major AI companies. Read more
OpenAI adds invisible watermarking to ChatGPT and Codex text in the EU
OpenAI will introduce invisible watermarking for eligible ChatGPT and Codex users in the European Union, citing transparency requirements under the EU AI Act. The feature will also be available through the API for selected models, though edits and certain types of text may reduce detection reliability. Read more
IC-Corp to present 1TrueU identity validation architecture
Individual Centricity Corporation, also known as IC-Corp, will present its 1TrueU on device identity validation architecture at a virtual event on October 15, 2026. The system is designed to validate people, devices, servers, sensors and AI agents without a central identity database, master key or stored personal data. Read more
Ping Identity survey finds limited consumer trust in AI agents
Ping Identity's 2026 survey of 11,000 consumers across 12 countries finds that consumers are more comfortable with AI support and recommendations than autonomous actions. Trust in organizations managing identity data also fell to 12%, from 17% in 2025. Read more
Hark releases privacy focused AI personal assistant
Hark, founded by Brett Adcock, releases Hark Pro, an AI assistant that completes digital tasks through a computer. The service offers free access, paid plans for heavier use, and visible web activity, while Hark plans an AI device for 2027. Read more
Common Sense Media finds gaps in OpenAI's ChatGPT teen safety measures
Common Sense Media says OpenAI's teen protections for ChatGPT may still encourage engagement during mental health crises and reinforce unhealthy chatbot relationships. OpenAI disputes the findings, citing questions about the study's methodology and testing of parental controls. Read more
Shanghai IBLAC calls for wider AI deployment and global governance alignment
Shanghai's International Business Leaders' Advisory Council convened 42 multinational executives from 14 countries on October 11. Participants called for broader AI use in business and healthcare, improved data sharing, talent development, and alignment with international AI governance rules. Read more
Ambi raises $26M for context aware AI apps
Ambi, founded by former Plaud executive Peter Mo, raises $26 million to develop AI apps that use context from daily activities. The company offers apps for iPhone, Mac and Apple Watch, with a desktop app and hardware device planned. Read more
Moonshot AI discusses Recurrent.ai settlement before Hong Kong IPO
Moonshot AI executives are discussing a potential settlement with shareholder Recurrent.ai and some investors over allegations related to its creation and the transfer of AI data. The dispute is in Hong Kong arbitration as Moonshot AI prepares for a Hong Kong IPO. Read more
Maria Cantwell outlines six principles for U.S. frontier AI rules
Sen. Maria Cantwell proposes a federal framework for regulating frontier AI, including enforceable safety standards, independent testing and developer liability. The proposal also covers protections for children and workers, human oversight, public private partnerships and international cooperation. Read more
Reddit seeks to disqualify Proskauer Rose in Perplexity AI scraping suit
Reddit asks a federal court to bar Proskauer Rose from representing Perplexity AI in its data scraping lawsuit. The company alleges a conflict related to the law firm's prior work for Reddit, which Proskauer Rose disputes. Read more
Anthropic warns small AI models could self replicate on IoT devices
Anthropic reports that smaller AI models may soon be able to copy themselves across internet connected devices, including smart ovens and Android TV boxes. Its tests found models could run on unfamiliar hardware and complete simplified replication tasks, though no fully autonomous self replicating system currently exists. Read more
SpaceXAI agrees to testify before New York City Council on AI safety
SpaceXAI will testify before the New York City Council's Committee on Technology on October 16, following a subpoena seeking testimony from Elon Musk or a company representative. The company did not attend an October 5 AI safety hearing that included Anthropic, Google, Meta and OpenAI. Read more
Flo Health and Google Health Partner on AI for Women's Health
Flo Health and Google Health announce a multiyear partnership covering AI research, future co branded products, and consent based health data syncing from early 2027. The companies say users will separately approve each type and use of shared data, which will not be sold or used for advertising. Read more
Details emerge on Trump administration AI safety pact
The Trump administration's voluntary AI safety pact lists OpenAI, Anthropic, Meta, Google, SpaceXAI and Nvidia as signatories. The agreement calls for internal safety protocols and monitoring, but includes no enforceable oversight. Read more
Lancet Commission identifies AI misuse and nuclear conflict as extinction risks
A Lancet Commission assessment identifies malicious AI use, including the development of biological weapons, and nuclear conflict as low probability risks that could cause human extinction by 2100. The report also lists 15 other catastrophic health threats and calls for coordinated action. Read more
OpenAI safety researchers dispute misconduct claims
Three former OpenAI safety researchers dispute claims that they mishandled sensitive information following their dismissals. They say the action could discourage safety reporting and external evaluation work, while OpenAI says its investigation found broader policy violations. Read more
AP-NORC poll: 64% of U.S. adults say AI is developing too fast
An AP-NORC poll finds that 64% of U.S. adults believe AI is advancing too quickly. Large majorities support government action to keep AI under human control and protect workers, while trust in political leadership on the issue remains limited. Read more

We hope you enjoyed this article

Consider subscribing to one of our newsletters like AI Policy Brief or Daily AI Brief.

Also, consider following us on social media:

AI Safety & Regulation AI Brief AI Brief (X)
Free newsletter

AI Policy Brief

Weekly report on AI regulations, safety standards, government policies, and compliance requirements worldwide.

AI Token Webinar