The FTC is scrutinising OpenAI and Anthropic, India is 30 days away from an AI rulebook, and Meta just shipped its first frontier model.
FTC sends 50-page demands to OpenAI and Anthropic
The US Federal Trade Commission has subpoenaed both OpenAI and Anthropic, requesting extensive records on security incidents, safety safeguards, marketing practices, and privacy. The 50-page demands signal a White House strategy of aggressive regulatory oversight without waiting for new legislation.
Why it matters: This is the most concrete regulatory pressure US AI companies have faced. It means AI safety is no longer just a self-policing promise — regulators are now asking for receipts.
India's AI rulebook is 30 days away
IT Minister Ashwini Vaishnaw announced on October 8 that MeitY will publish a consultation paper on AI regulation within 30 days. The paper will cover deepfakes, AI safety, and a "techno-legal" approach pairing rules with technical measures. Alongside it, the government is procuring 10,000 additional GPUs for public-interest AI workloads.
Why it matters: India has moved from watching other countries regulate AI to setting its own agenda. If you build AI products for Indian users, the compliance clock just started ticking.
Meta launches Muse: its first frontier personal AI agent
Meta released Muse Wonder, described as the company's first frontier model, alongside Muse — a personal AI agent that runs on a dedicated virtual machine in Meta's cloud. Unlike chatbots you summon, Muse is designed to work on tasks in the background on your behalf.
Why it matters: Meta entering the frontier model race — not just fine-tuned open-weights territory — changes the competitive picture. A billion-user platform with its own frontier model is a different proposition from any current AI assistant.
Claude Haiku 5.5: effort controls and 1M context in a small model
Anthropic's Claude Haiku 5.5 is now available, marking the first Haiku with effort controls — letting developers tune how much the model "thinks" before replying — plus a 1 million token context window. It is designed to bring frontier-class reasoning to cost-sensitive deployments.
Why it matters: Effort controls in a small, cheap model means developers can build smarter apps without paying frontier model prices for every query. This closes a practical gap that previously forced an expensive trade-off.
UK's AI Security Institute is testing Gemini 4 Argon
Britain's AI Safety Institute has confirmed it is evaluating Google's Gemini 4 Argon, coordinating with the US government and major AI developers. The institute is running independent safety checks on the latest generation of models before they reach consumers.
Why it matters: Pre-deployment safety testing by a government body — not just the developer — is becoming standard practice. The era of models launching with only the company's own safety sign-off may be ending.
UAE releases Jais 3: six Arabic language models up to 375B parameters
Abu Dhabi's MBZUAI and Inception42 launched Jais 3, a family of six open Arabic language models ranging from 900 million to 375 billion parameters. It is one of the largest open-model releases targeting Arabic-language AI specifically.
Why it matters: Most frontier models are built in English and ported to other languages. A 375B open Arabic model changes what is possible for the 400 million Arabic speakers who have been underserved by current AI tools.
Anthropic's IPO valuation is being marked down
Investor day discussions at Anthropic are reportedly centred on a $1.5 trillion IPO valuation, down from earlier estimates of about $2 trillion. The gap matters because it shows how the market is beginning to separate AI companies with real revenue from those still running on promise. Where Anthropic lands when it goes public will set a reference point for the entire sector.
Microsoft calls AI "less regulated than milk" + Anthropic's AI welfare rule + Siri goes multilingual
Microsoft President Brad Smith renewed calls for global AI safety rules, saying AI is currently less regulated than milk. Anthropic updated its usage policy to ban sustained abusive behaviour toward Claude. And Apple's iOS 27.2 — expected before October ends — will bring Siri's AI features to French, Japanese, Korean, Portuguese, and Spanish for the first time.