The Omniscient Bulletin · October 16, 2026
The FTC reportedly serves safety demands on two labs as investors reportedly trim Anthropic's IPO valuation
The FTC's inquiry into OpenAI and Anthropic, first reported in September, reached written demands on Wednesday, according to one source who spoke to Axios. Investors are pricing AI risk more carefully: some discussed about $1.5 trillion for Anthropic's IPO, down from roughly $2 trillion, and CoreWeave's debt traded at distressed levels. Meta released a frontier-scale open-weight model, and Microsoft detailed one it has finished pre-training.
Policy
The FTC sent OpenAI and Anthropic civil investigative demands over recent AI safety incidents, a source told Axios
The Federal Trade Commission on Wednesday sent civil investigative demands to OpenAI and Anthropic seeking information on recent safety incidents, a source told Axios. The source said the requests ran to more than 50 pages and sought documents on the incidents, internal security measures and marketing materials, and that third parties targeted by the companies' agents will be subpoenaed later. Reuters reported in September that the probe was coming. Neither company nor the FTC responded to Ars Technica.
Frontier
Meta released the weights for Muse Spark Open, an 816 billion parameter model it calls the most capable US open-weight model to date
Meta released the weights for Muse Spark Open, which it calls "the most capable US open-weight model to date": a multimodal model with 816 billion total parameters, 42 billion active, and a 1 million token context window, built for coding and agentic work. Meta says no model in its class has run on a desktop before; this one runs on two Apple M5 Ultra Mac Studios. The license has limited commercial terms. Meta says it met its safety framework's release thresholds, with a full evaluation report available soon.
Industry
Some Anthropic investors discussed a valuation of about $1.5 trillion for its IPO, down from roughly $2 trillion, Reuters reported
Some attendees of Anthropic's Wednesday investor day in San Francisco talked about valuing the company at around $1.5 trillion for its November share sale, down from the roughly $2 trillion investors had expected, three people told Reuters. Some were disappointed by slowing sequential growth: third-quarter revenue rose 45% to $16.8 billion after more than doubling in the second quarter. The Information reported earlier that investors were weighing a lower valuation. Anthropic did not immediately respond.
Industry
Anthropic gave six METR staff employee-like access to its systems and says it cannot veto what METR publishes
Anthropic said it has onboarded six employees of METR, a nonprofit that tests frontier AI models, with company hardware, office space and internal tools roughly on par with its own risk assessment teams. METR will publish its analysis of Anthropic's risk levels, incidents or practices; Anthropic cannot veto publication but can require redactions in narrow, predefined categories. SecureBio and Accenture are evaluators too, in earlier-stage partnerships, and Anthropic is developing three tiers of evaluator access.
Compute
Jensen Huang said Nvidia sees no need to backstop CoreWeave, as its debt traded at distressed levels
Nvidia CEO Jensen Huang told reporters in Stockholm that he sees no need for Nvidia to backstop CoreWeave, which has about $50 billion in debt. CoreWeave's April-issued bonds fell to 79.15 cents on the dollar, yielding around 15%, above the level that typically signals distress, and the cost of insuring its debt reached around 1,000 basis points, which Reuters says implies close to a 63% probability of default within five years. Nvidia has agreed to buy any unsold CoreWeave capacity through April 2032.
Frontier
Microsoft AI finished pre-training MAI-Base-2, a 2.8 trillion parameter model it says beats every open-weight model it tested
Microsoft AI said it pre-trained MAI-Base-2, a mixture-of-experts foundation model with 2.8 trillion parameters in total, 80 billion of them active, on 16,000 GB200 GPUs across 38 days in August and September. It says the model outperforms every open-weight model it tested on its own held-out evaluations and is 4 to 12 times more efficient in pre-training compute than DeepSeek-V4-Pro-Base. The model is now in post-training with no release date, and the benchmark results are Microsoft's own.
Policy
The US Treasury is pressing IMF partners to call AI 'super intelligence' in a statement now under negotiation
The US wants other governments to start calling artificial intelligence "super intelligence," Treasury Undersecretary Erin Browne told Reuters in Bangkok. Washington is negotiating with other IMF shareholders to use the term in a statement from the fund's steering committee, and as G20 president it wants the change adopted internationally. Browne declined to say whether the US would refuse to sign off without it; a source said US officials suggested they would insist. The statement is due Saturday.