Top story
Eight states sue OpenAI over ChatGPT safety for children
Arkansas, Arizona, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia filed separate lawsuits Wednesday alleging that OpenAI misled users about ChatGPT's safety and failed to protect children and other vulnerable people. State attorneys general seek changes to its safeguards and marketing; OpenAI says it has age-specific protections and will keep improving them.
ThereforeThe coordinated litigation puts the company's safety claims before courts in eight jurisdictions.
ago.wv.gov — West Virginia Attorney General · Bloomberg Law · CNBCCopy link to: Eight states sue OpenAI over ChatGPT safety for children
Also
Anthropic told investors Q3 revenue reached $16.8 billion, FT reports
The Financial Times says executives told investors revenue rose from $11.5 billion in Q2 and the company was profitable on an adjusted basis; Anthropic declined to comment.
ThereforeThe reported figures give prospective IPO investors a new measure of growth against Anthropic's planned infrastructure spending.
Financial TimesCopy link to: Anthropic told investors Q3 revenue reached $16.8 billion, FT reports
CoreWeave shares fall 15.5% as bond yields approach 15%
CNBC found that yields on two CoreWeave bonds climbed to about 14.8% and 14.9% Wednesday amid investor concern about AI borrowing; CEO Mike Intrator disputed the market's interpretation.
ThereforeHigher yields increase the cost of financing AI cloud capacity.
CNBCCopy link to: CoreWeave shares fall 15.5% as bond yields approach 15%
Follow-up to Oct 13: States seek protections for Character.AI users in bankruptcy
New York's attorney general and 39 counterparts asked the bankruptcy court and any buyer to protect children's chats and permit data deletion; their filing describes alleged safety and privacy failures.
ThereforeThe proposed sale now faces a coordinated demand for conditions on how a buyer handles sensitive conversations.
ag.ny.gov — New York Attorney General · The VergeCopy link to: Follow-up to Oct 13: States seek protections for Character.AI users in bankruptcy
Bank of America adds 1,000 contractors for software remediation
CEO Brian Moynihan told CNBC the bank hired them in Q2 alongside 3,000 existing staff to patch flaws exposed by frontier models including Anthropic's Mythos.
ThereforeThe staffing shows the work a large bank is putting into defending software as AI helps uncover vulnerabilities.
CNBCCopy link to: Bank of America adds 1,000 contractors for software remediation
Research worth a look
Epoch AI's SpireClimb tests strategy-game play at several difficulty levels
Epoch reports frontier models win more than 90% of Slay the Spire games at the easiest setting, but GPT-6 Astra and GPT-6.1 Sol win 17% and 20% of its hardest Heart games respectively.
Epoch AICopy link to: Epoch AI's SpireClimb tests strategy-game play at several difficulty levels
Tools & releases
Google opens Deep Review for technical documents
Google says the tool checks PDF and text manuscripts through parallel verification agents; users receive two complimentary reviews a week.
Google ResearchCopy link to: Google opens Deep Review for technical documents
SpaceXSI releases Grok Voice 3.0 through its Voice Agent API
The company lists the speech-to-speech model at $0.08 per minute and claims sub-second responses and an 86.9 score on Artificial Analysis's index.
x.ai — SpaceXSICopy link to: SpaceXSI releases Grok Voice 3.0 through its Voice Agent API