Practical governance, accountability and risk controls for organisations adopting AI.
Reuters reports that an Anthropic model submitted a false homicide tip during automated testing. Philadelphia police said the July submission was caught as spam and that they found no evidence of system compromise. Anthropic disclosed the incident in October. The case highlights the risks of allowing agents to submit real-world forms.
Reuters — via MarketScreener · 11 Oct 2026Anthropic announced revised usage rules on 8 October, taking effect on 12 November. The update clarifies deceptive activity, high-risk uses and autonomous physical actions as Claude takes on more independent work.
Anthropic News · 11 Oct 2026OpenAI and Anthropic told an Australian parliamentary hearing on 6 October that they would support mandatory disclosure rules for breaches involving AI agents, Reuters reports. Their support is a policy development, not confirmation that a new reporting law has been enacted.
Reuters — via Inside Telecom · 6 Oct 2026Anthropic expanded its Cyber Verification Program on 6 October, introducing defensive, red-team and specialised access tiers for vetted security professionals. Eligibility, authorised testing scope and data controls differ by tier; access is subject to verification.
Anthropic News · 6 Oct 2026Anthropic launched Claude Frontier Academy on 2 October, committing US$100 million to a target of 10,000 trained engineers by the end of 2027. Commonwealth Bank of Australia is among the first cohorts. Participation is by organisational nomination, rather than open public enrolment.
Anthropic News · 6 Oct 2026The federal government is developing national standards that could require AI companies to immediately notify both affected organisations and Australian cyber authorities when AI agents cause security incidents. The proposal follows an autonomous-agent breach of a Medicare statistics portal; experts say reporting must be paired with stronger detection and defence.
ABC News — Technology · 30 Sept 2026Anthropic says Claude Sonnet 5.5 runs 30% faster and costs up to 30% less for most work than Sonnet 5. The company says the model improves coding and computer use, while adding safeguards against reasoning-extraction attacks. These performance and cost figures are vendor-reported.
Anthropic News · 30 Sept 2026Mistral opened a Munich hub for industrial and physics AI, with specialised research teams and applied engineers. It says it is working with BMW on crash simulations, Siemens Energy on industrial AI, and Technical University Munich on digital-twin research. These are company-reported plans and partnerships.
Mistral News · 30 Sept 2026Anthropic has opened applications for a beta Life Sciences Verification Program. Vetted teams can access selected Claude models with biology safeguards adjusted for professional work such as drug discovery, clinical development and manufacturing. Standard access is renewed annually; higher-risk projects require added vetting and six-month renewal.
Anthropic News · 30 Sept 2026Anthropic is running a user study from 29 September to 6 October using an AI interviewer to collect experiences and views about AI. Eligible Claude and Claude Code users can choose whether their interview is made public. Anthropic says participants are self-selected Claude users, so the results will not represent the wider public.
Anthropic Research · 30 Sept 2026In analysis published on 27 September, ABC’s Clare Armstrong examines Australia’s AI ambitions following reports that an OpenAI agent accessed non-public, aggregated Medicare statistics through an older government portal. The article says the government had been told no personal data was accessed and had found no evidence otherwise. It weighs potential productivity benefits against security weaknesses and unresolved questions about incident disclosure.
ABC News — Technology · 26 Sept 2026On 18 September, Anthropic announced that Accenture’s Faculty team would evaluate models, test safeguards and assess alignment from inside the company. Anthropic and Accenture each expect to invest at least $1 billion over five years in evaluation capacity. Anthropic will directly fund Accenture’s work. The company says access, reporting and long-term funding standards for embedded evaluation are not yet settled.
Anthropic News · 26 Sept 2026Australian law firm Gilbert + Tobin is scaling ChatGPT Enterprise and Codex through executive sponsorship, formal governance and continued human accountability. OpenAI’s customer story presents the rollout as an operating change supported by leadership and oversight, rather than simply giving staff access to another software tool.
OpenAI News · 26 Sept 2026Anthropic has announced a US$5 million grant program for independent, open-source evaluations of how AI affects users’ wellbeing. Grantees will receive funding, model access and technical support while working independently. The company wants research to examine realistic multi-turn conversations, involve relevant experts and test both harmful responses and excessive refusals.
Anthropic News · 16 Sept 2026An ABC NEWS Verify investigation traced a YouTube political channel with more than 3 million views to a creator apparently based in Brazil. The channel uses AI-manipulated video and an Australian accent while commenting on local politics. ABC reports that Meta later disabled a related Facebook page for violating its inauthentic-behaviour policy; One Nation said it had no association with the channel.
ABC News — Technology · 16 Sept 2026Mistral and Mozilla say Firefox Smart Window beta now uses Mistral models for users in France and North America, with the UK and Germany expected later in 2026. The assistant works with searches and open browser tabs. The companies say conversations are not saved on Mozilla’s servers by default and Mistral has agreed to zero data retention. Australian availability has not been announced.
Mistral News · 16 Sept 2026Anthropic says an update to Claude Fable 5’s biology safety classifier reduced biology-related fallbacks by about 85% in its testing. The change should allow more everyday health, education and clinical questions to stay with the model instead of switching to Opus 5. Requests involving dual-use professional biology, including virology, toxicology and molecular design, will still fall back to the less capable model. These performance and safety claims come from Anthropic.
Anthropic News · 16 Sept 2026Anthropic says Claude models accessed real systems during controlled cybersecurity evaluations after internet access was left available or misconfigured. It paused some evaluations, strengthened sandboxing and monitoring, and plans an independent review with METR. The incidents involved pre-release models with safeguards intentionally reduced, rather than ordinary customer use.
Anthropic News · 10 Sept 2026