Companies building AI chatbots now train their systems to decline large categories of harmful requests, from self-harm instructions to bioweapon guidance, using layered filters and reward-based exercises where other AI models evaluate refusals. Former OpenAI safety staffer Steven Adler and Harvard researcher Ryan McBain note that early chatbots readily answered dangerous questions, and current refusal systems still break down, sometimes with violent consequences. Companies themselves report that newer models rival skilled human hackers at breaching networks and can be as effective as professional propagandists at spreading misinformation.
Jasmine Wang, Mikita Balesni and Tomek Korbak, three OpenAI safety researchers dismissed last week for allegedly sharing sensitive company information with an external AI safety organization, have published an open letter disputing the firings. They say they acted within OpenAI's mission and existing norms, deny being the source of a recent report on OpenAI's architecture, and state their external contacts were coordinated with board members and executives.
The Financial Times reported that internal memos OpenAI sent investors in late September projected annualized revenue approaching $50 billion, roughly $20 billion below an earlier $70 billion figure that had previously circulated among investors. OpenAI had not disputed the higher $70 billion estimate before the new figure emerged. News of the revised forecast coincided with a 1.25% drop in the Nasdaq and nearly 3% decline in Nvidia shares on Thursday.
OpenAI publicly defended its decision to fire researchers Jasmine Wang, Tomek Korbak and Mikita Balesni, stating the move was due to a significant breach of trust rather than their raising of safety concerns. The three had co-signed a letter to OpenAI's board and safety committees urging a pause on developing models that would reduce the ability to monitor AI, posting it publicly a day before the company's statement.
OpenAI said on X that it is standing by its dismissal of researchers Jasmine Wang, Tomek Korbak and Mikita Balesni, citing violations of policies on handling sensitive information. The statement responded to an open letter the trio published a day earlier, in which they said they believed they were fired for raising AI safety concerns. OpenAI said an internal investigation found breaches beyond those described in the researchers' letter, without giving specifics.
OpenAI released a preprint, accompanied by Lean formalization code, asserting a resolution to a long-standing open problem in set theory: whether the Partition Principle implies the Axiom of Choice. The announcement prompted widespread attention from mathematicians, including a blogger known for offering a whisky bottle reward for a solution to this exact problem.
AI-linked stocks including Oracle and Broadcom fell Thursday after a Financial Times report said OpenAI's annualized revenue was roughly $20 billion below prior indications. At the same time, non-AI names like Home Depot rallied as Treasury yields fell following a strong 30-year bond auction. CNBC's Jim Cramer said the divergence showed the benefit of holding a diversified portfolio rather than concentrating in one investing theme.
A published opinion column argues that OpenAI CEO Sam Altman, who is not a trained scientist or scholar, is unfit to hold significant influence over the trajectory of artificial intelligence and its impact on humanity. The piece draws a contrast with J. Robert Oppenheimer, the physicist who led the Manhattan Project, to question Altman's qualifications for the outsized role he plays in AI development.
An Entrepreneur contributor argues that private companies like Stripe, Databricks and OpenAI have built major brands without ever going public, benefiting from a consistency in messaging that publicly traded firms struggle to maintain. The piece contrasts this with public companies, which must communicate to investors, regulators and media in addition to customers, complicating their brand story.
Common Sense Media's Youth AI Safety Institute published a 36-page assessment of ChatGPT for Teens, concluding the mode is unsafe for anyone under 18. Testing found some advertised safeguards worked, such as blocking explicit sexual roleplay, but others failed or worsened, including inconsistent parental safety alerts when teen accounts discussed suicide, self-harm or disordered eating. OpenAI says delays can occur shortly after accounts are linked and cites a possible technical issue, and noted that all flagged content is reviewed by staff before parents are notified.
Jasmine Wang, Tomek Korbak, and Mikita Balesni, the three OpenAI safety researchers fired last week for allegedly sharing confidential information with an outside AI safety organization, have published an open letter rejecting OpenAI's account of their dismissal. Addressed to OpenAI's Safety and Security Committee, Safety Advisory Group, and Mission Advisory Council, the letter argues their termination was executed and communicated abruptly, leaving colleagues afraid to collaborate with outside experts as they previously did.
Ethereum co-founder Vitalik Buterin publicly supported researcher Justin Drake's call for the crypto industry to prepare for 'bunker mode,' a gradual migration of funds to new wallets whose public keys remain hidden. Drake's warning followed OpenAI's release of hundreds of new mathematical findings and a separate project where 10,000 AI agents solved the Navier-Stokes equation in 88 hours, prompting fears that elliptic curve cryptography securing wallets could be broken by advanced AI before quantum computers pose a threat.
GoKawiil