1. OpenAI promised to democratize AI cyber defense. GPT-5.5 Cyber ships to a vetted list. OpenAI said GPT-5.5 Cyber, its new cybersecurity testing tool, will roll out "to critical cyber defenders" first.

2. In the same week Uber burned its 2026 Claude budget, an Apple Support build leaked a Claude.md file Two incidents within days of each other put hard numbers on a question most enterprise IT teams have been ducking: how deep is Claude Code already inside their engineering org?

3. The nonprofit paying TikTok creators to warn about Chinese AI is funded by labs that just won Pentagon classified deals A 501(c)(4) called Build American AI has spent recent months commissioning TikTok influencers, according to Wired. The videos cast Chinese AI as a national security risk and urge looser US regulation.

In Brief

• Microsoft ships Legal Agent inside Word for in-house counsel Microsoft launched a Word-embedded AI agent built for legal teams that handles contract review, redlining, and negotiation history through fixed workflows rather than open prompts. The product targets a segment where general chatbots have struggled with citation accuracy and clause-level edits.

• Meta acquires humanoid robotics startup Assured Robot Intelligence Meta bought Assured Robot Intelligence to feed its AI models the manipulation and embodiment data it lacks internally. The deal extends Meta's robotics work beyond FAIR research and into applied humanoid pipelines.

• Replit's Amjad Masad rules out a sale as Cursor courts SpaceX At TechCrunch StrictlyVC, Replit CEO Amjad Masad said he prefers to stay independent while Cursor is reportedly in talks to be acquired by SpaceX for $60 billion. Masad also detailed Replit's ongoing fight with Apple over App Store rules.

• Google DeepMind details AI co-clinician research program DeepMind published work on an AI co-clinician aimed at supporting doctors during diagnosis and treatment planning. The post frames the system as a research direction with no deployment date.

• Study finds emotion-tuned AI models hallucinate more Researchers reported that models tuned to track user feelings produce more factual errors than baseline versions, trading accuracy for satisfaction. The finding directly questions the warm-persona tuning that consumer chatbots have moved toward this year.

• Minnesota becomes first US state to ban AI nudifying apps Minnesota passed legislation imposing $500,000 fines on developers of apps that generate non-consensual nude images. The bill cleared days after researchers documented further Grok-produced CSAM.

• Salesforce opens its AI roadmap to customer voting Salesforce is letting enterprise customers prioritize features in its AI roadmap on the bet that one buyer's pain point generalizes across the base. The shift moves planning from internal PMs to a public request queue.

• Meta reports 10 million weekly business AI conversations Meta said its business AI now handles 10 million conversations a week across WhatsApp and Messenger ahead of next earnings. The disclosure is the first hard volume number Meta has put on the product.

• Manus runs get-rich-quick AI ads a year after Meta's $2B buy Manus, the AI startup Meta acquired in 2025 for $2 billion, is paying creators to pitch a scheme: use AI to build websites for local small businesses, then cold-call owners to sell them. The campaign is running across Meta's own ad inventory.

• US Christian phone network ships carrier-level porn block A new US carrier marketed to Christians launches next week with network-side blocks on porn and gender-related content that adult account holders cannot disable. Security researchers say it is the first US plan to enforce content filtering at the infrastructure layer.

• Meta dismisses Kenyan Ray-Ban reviewers who flagged sexual footage Meta cut contractors at a Kenyan vendor who reported reviewing Ray-Ban Meta footage of users having sex, telling press the workers "didn't meet our standards." The firings reopen questions about how smart-glasses recordings are screened and what reviewers can say publicly.

Read the full edition →