5 stories about Safety and trust, newest first. Each one links to its primary source.
On 5 October 2026 OpenAI said it will add an invisible statistical watermark to eligible ChatGPT and Codex text output in the European Union over the coming weeks. API customers worldwide can opt in from 5 October; it is off by default. OpenAI says text watermarking has significant limitations.
On 8 October 2026 Anthropic announced an updated Usage Policy that takes effect on 12 November. It adds safety requirements for Claude controlling equipment, bars Claude from deciding who to investigate, arrest or charge, and adds a ban on sustained, needless abuse of its models. Anthropic says most changes clarify existing rules.
On 8 October 2026 OpenAI reported banning two covert influence operations that used its models: a Russia-origin cluster it calls Dark Clark, and an Iran-origin one it calls Bogus Bylines. OpenAI rates Dark Clark Category 5, the first it has disrupted since it began reporting. The findings and ratings are OpenAI's own.
On 7 October 2026 Google made its SynthID Detector available to anyone, globally and in English. The tool checks whether an image, video or audio file carries a SynthID watermark from Google or partners named as OpenAI, NVIDIA and Kakao; Apple is listed as coming soon. The post does not describe limits, so a missing watermark proves nothing.
Between 28 September and 5 October 2026, OpenAI launched Dots agents and GPT-6.1 Sol but scrapped GPT-6.1 Astra after safety tests. Google released Gemini 4 Argon to cybersecurity partners only, six AI companies signed a voluntary White House accord, and Anthropic's IPO filing was reported.