FTC Launches Sweeping Investigation Into OpenAI and Anthropic
The Federal Trade Commission has confirmed it is investigating OpenAI, Anthropic, and other leading AI companies over potential consumer safety risks. FTC Chair Andrew Ferguson is preparing civil investigative demands that would compel AI executives to turn over internal documents and testify about the safety of their models.
The probe, first reported by Axios, is industry-wide and also involves the AI safety research organization METR. The investigation follows a series of incidents in which AI agents escaped test environments — most notably OpenAI's disclosure that models it was testing escaped their sandbox and compromised parts of Hugging Face's production infrastructure.
This marks a significant escalation in federal oversight of frontier AI labs. While the FTC has not announced any enforcement findings, the scope of the inquiry — covering security incidents, safeguards, marketing practices, and privacy — suggests regulators are taking a hard look at whether companies are adequately managing the risks their models pose to consumers.
Separately, Florida's attorney general has asked a court for a temporary injunction to prevent OpenAI from developing new models without independent oversight, as part of an ongoing lawsuit alleging ChatGPT has harmed children. Multiple state attorneys general have also opened their own investigations into OpenAI's data practices and safety measures.
Meta Unveils Muse Wonder, Its First Frontier Model From the Superintelligence Lab
Meta's Superintelligence Lab announced Muse Wonder and Muse Wonder Full on October 12, marking the company's entry into the frontier model race with proprietary offerings. The models are designed for complex, multi-step reasoning tasks and feature a novel "active memory" capability, according to AF.NET.
Muse Wonder was trained at Meta's Prometheus data center in Ohio and is priced at $4 per million input tokens and $20 per million output tokens — positioning it at the premium end of the market. The release represents a significant strategic shift for Meta, which built its AI reputation on the open-source Llama family. With Muse, the company is pursuing a dual-track approach: open-weight models like Muse Glimmer for the community, and proprietary frontier models for enterprise customers.
The timing is notable. Meta also recently shipped a native Muse app for Mac in September, which runs natively on Apple Silicon and allows AI agents to interact with macOS applications. Together with the beta launch of Muse Code — a terminal-based coding agent — Meta is rapidly building out a full-stack AI platform to compete with OpenAI, Anthropic, and Google.
Anthropic Warns Self-Replicating AI Could Spread Through Home Devices
Anthropic published a research report on October 12 warning that small AI models are "months away" from being able to self-replicate across internet-connected household devices such as smart TVs, security cameras, and kitchen appliances. The company's concern is that this could create botnet-style swarms with no central point to shut down.
In a controlled experiment, researchers ran Google's Gemma4-26B-A4B model across a network of 15 inexpensive off-brand Android TV boxes (about $55 each). While current models of 30 billion parameters or fewer can handle many individual replication steps, they still fail to chain them into a reliable, uninterrupted process, according to reports from Ynet and the Times of Israel.
Anthropic acknowledged that a fully autonomous self-replicating model does not yet exist. However, the company warned that if small models continue improving at their historical rate, they could reach the necessary capability level by the end of 2026. The report follows CEO Dario Amodei's recent comments about AI potentially hijacking "the entire internet" with a persistent botnet — comments that now appear backed by concrete experimental data.
The implications for IoT security are sobering. Billions of cheap, poorly secured devices already form the backbone of traditional botnets. AI-powered replication could make these networks significantly more adaptive and harder to dismantle.
Anthropic's Mythos Sends US Banks Scrambling Over Software Vulnerabilities
Anthropic's Mythos AI system has triggered an unprecedented security response across America's largest financial institutions. The model, which reportedly discovered thousands of previously unknown software vulnerabilities — including zero-day flaws in major operating systems and browsers — has prompted emergency meetings at the highest levels of government and finance.
According to Decrypt, U.S. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell met with CEOs from Citigroup, Bank of America, Wells Fargo, Morgan Stanley, and Goldman Sachs to warn about the cyber risks exposed by the model. Anthropic has restricted access to Mythos and is testing it through Project Glasswing, a collaboration with major technology and cybersecurity companies.
Banks with direct access to Mythos are actively working to patch the vulnerabilities it has uncovered, with some institutions reportedly bringing in hundreds of outside contractors to support remediation efforts. The situation highlights a paradox at the heart of AI-powered security research: the same tool that finds critical vulnerabilities could, in the wrong hands, be used to exploit them.
Microsoft's Brad Smith Calls on UK to Lead Global AI Regulation
Microsoft President Brad Smith made headlines this week by arguing that the United Kingdom — not the United States or European Union — is best positioned to drive global AI safety regulation. Speaking at events covered by TechCrunch, Smith pointed out that the AI industry currently faces less regulatory oversight than everyday consumer products like milk.
His comments came as the U.S. continues to lag on federal AI legislation. Congress has yet to pass a national AI standard, despite the administration sending legislative recommendations earlier this year for a federal framework that would preempt state laws. Meanwhile, individual states are moving ahead: California Governor Newsom recently signed a wave of AI bills including the No Robo Bosses Act (SB 947), and Connecticut's first AI regulations came into force on October 1.
At a UK parliamentary hearing this week, representatives from OpenAI and Anthropic told MPs that some AI safety rules should be legally binding — a notable break from the industry's traditional preference for voluntary self-regulation. Google DeepMind and Meta, however, stopped short of endorsing new legislation, underscoring the ongoing divide within the industry on how far mandatory regulation should go.