OpenAI and Anthropic this week: Astra for Law, misalignment reports, one Claude, Hacktron, standards body (Week 38, 2026) Hacktron hacked OpenAI in late July under OpenAI's bug bounty safe harbor, using Claude Opus 5 to exploit a Discourse bug and grab ChatGPT and Codex sign-in tokens used to reach Slack, GitHub and the internal Monorepo, per Hacktron and the WSJ 404 Media reports hundreds of OpenAI contractors read real ChatGPT conversations under Project Lily, and Anthropic confirmed they also use human review for users who turned on Help improve our AI models OpenAI introduced Astra for Law on GPT-6 Astra with a Legal Search Index of more than 230 million URLs, initially for selected firms GPT-5.3-Codex-Spark is gone, and GPT-5.5 leaves ChatGPT and Codex on October 14 but stays in the API OpenAI published a framework for disclosing model misalignment, writing they do not believe alignment and monitoring are solved well enough to keep responsibly scaling at maximum speed for much longer Anthropic is merging Claude Cowork and chat into one Claude, rolling out to Pro and Max over the next few weeks, and decks, docs and designs can now be made inside a conversation Anthropic launched Salesforce in Claude in beta, Claude for Financial Advisors, and 27 new Claude for Small Business integrations Anthropic's Life Sciences Verification Program opens Mythos 5.1 for biology work for the first time, on Enterprise, Team and the Console Anthropic redesigned Projects in Claude Code as one conversation that splits work into parallel cloud threads, in beta for select Pro and Max users Anthropic and Accenture each expect to invest at least 1 billion dollars over five years in frontier AI evaluation capacity, and more