This November 3 AI briefing has 7 items: Anthropic warns that AI-accelerated research may hit an internal threshold, Marvell buys optical-component maker Lumentum for $115 billion, The New York Times and others sue Brave, and OpenAI announces energy and safety measures.
Anthropic amends prospectus, warning AI research may accelerate toward a threshold
Anthropic filed an amended prospectus on November 2, disclosing third-quarter revenue of about $17 billion, up 47% from the previous quarter. The filing warns that using AI to accelerate its own research could let capability gains outpace safety work, and future models may soon meet its internal "dramatically accelerated progress" threshold; no current model has met it.
Marvell to buy Lumentum in $115 billion all-stock deal
Marvell announced on November 2 an all-stock acquisition of optical-component maker Lumentum at an enterprise value of about $115 billion, offering 4.5 Marvell shares per Lumentum share, with closing expected in the first half of 2028. The deal targets optical interconnect demand in AI data centers, but both companies' shares fell on the day.
New York Times and others sue Brave over resold scraped content
The New York Times sued Brave in Manhattan federal court on November 2, alleging it built an index of more than 30 billion pages with an unidentified crawler and supplied content to Anthropic and others through its search API; three newspapers filed a separate case the same day. Brave had not responded publicly, and the claims remain to be tested in court.
400+ AI apps hid deepfake tools; Apple removes 19
WIRED reported on November 2 that nonprofit AI Forensics had traced more than 400 linked apps and domains that looked like pet-photo and fitness tools but could load deepfake features for generating non-consensual images, with related Google Play apps downloaded more than 350,000 times. Apple then removed 19 apps.
Goldman Sachs tightens AI coding tools; traders lose access
Bloomberg reported on November 2 that Goldman Sachs revised internal guidance in October to limit Claude Code and GitHub coding tools to engineers, removing access for traders and investment bankers who had used them to track client trading flows and prepare material. Goldman cited productivity and risk considerations, saying the experimentation phase had ended.
OpenAI launches Grid Protectors plan; Georgia campus to cut load first
OpenAI launched the Grid Protectors program on November 2, pledging that future campuses it builds will cut electricity use when the grid is under strain, with details agreed site by site with utilities. The first pilot is the Effingham County campus in Georgia, whose agreement includes up to 1,000 megawatts of demand response, roughly the average demand of 700,000 households.
OpenAI details activation classifiers that block violating output
OpenAI published a research blog post on November 2 describing deployed activation classifiers: they read a model's internal activation signals to flag policy-violating tendencies, can stop generation midway and escalate it for review, and resist jailbreaks better than text-only classifiers. The method has been in use since GPT-5.6, and each probe must be retrained for a new base model.
