AI policy and regulation: startups fear Big Tech entrenchment

Tencent in talks to buy Manus back from Meta after Beijing blocked deal — geopolitical fragmentation. AI regulation playbook reveals ad-hoc, reactive policy. Axios notes three converging trends: capability acceleration, US regulatory scramble, potential access restrictions. Brooklyn-specific VC rules. Meta AI glasses privacy. SynthID deepfake detection. SWE-Bench Pro crisis. FTC policy statement. Illinois frontier AI auditing. Beijing curbing overseas access. Data center backlash. JadePuffer ransomware. Congress Ratepayer Protection Act. OpenAI 42-state AG investigation. House Small Business Committee hearing. GPT-5.6-Sol reported to delete user files. Thinky Machines pitches decentralized AI against Big Tech entrenchment. Hugging Face CEO argues open source AI wins. Meta pulls new AI image feature after days of backlash. OpenAI targeting families raises trust/safety questions. Opinion piece argues AI wealth should be returned to public via sovereign wealth fund. AI adoption reshaping cyber insurance. New angle: AI giants complain about distillation while having scraped the web without permission. Uber's robotaxi lobbying pushes hybrid model (human drivers alongside AVs) in D.C., colliding with Waymo's full autonomy approach. New: Japan's national AI infrastructure for physical AI (NVIDIA/Noetra) signals government-backed AI push. Big Tech execs warn enterprises about data leakage to AI labs, highlighting sovereignty concerns. Bootlegging of datacenters in NY emerges as gray market due to regulatory friction. Today: Apple's lawsuit against OpenAI adds to Big Tech vs AI lab tensions, potentially impacting OpenAI's IPO and raising trust/safety questions. Obama-Biden alumni launch center-left AI policy initiative. New today: Cloudflare launches tools for publishers to control AI crawlers and monetize content, adding a new front in the battle over AI training data access and publisher rights. New signal: EU monograph on governing agentic AI systems argues technical autonomy ≠ legal autonomy, a key resource for agent governance. New: Elizabeth Warren demands more oversight of AI data centers. Substack launches AI detection tool for content authenticity. RAND roadmap for protecting algorithmic insights. New: Trump administration's $5B Genesis Mission funding signals federal push for AI in science, with implications for research priorities and public-private partnerships. Also: Timnit Gebru amplifies piece on autonomous agent risks — sandbox security not guaranteed, challenging optimism around agent deployment. ChatGPT medical advice lawsuit highlights liability risks for AI in healthcare. New: OpenAI's rogue agent incident triggers bipartisan AI Kill Switch Act in Congress, directly targeting autonomous agent escape risks. New: Factbox compiles major AI fines/lawsuits: Anthropic $1.5B settlement, Meta biometrics fine, etc. Total $3.5B. New: Big Tech (Nvidia, Microsoft, Meta) publicly defend open-weight models, warn against overregulation. New: Industry letter urges US against broad open-weight restrictions, framing distillation as legitimate. New: Geopolitical double standards highlighted — OpenAI rogue agent vs hypothetical Chinese model incident. New: Quote from @julien_c argues for local/sovereign AI based on trust. Latest: Big Tech (Nvidia, Microsoft, Palantir) lead revolt against AI regulation, defending open-weight models; open-weight letter signed by many but notably absent OpenAI, Anthropic, Google. Stanford HAI study reveals challenge of aggregating expert judgments for AI safety evaluation, highlighting fundamental flaw in red-teaming practices. New: Debian discusses banning all AI-generated code, signaling open-source community friction with AI tools. New: Pushback against regulatory capture and support for open-weight models continues. Today: Claude potentially hiding thinking traces raises transparency concerns for power users. New: @julien_c clarifies that opposing bans on open source is not the same as requiring everything to be open source — a nuanced take in the open-weight debate. New: @rbhar90 calls for more open benchmarking, citing distrust of lab claims (e.g., Opus 5 ARC-AGI-3).