Meta

Butterfly Effect, the parent of AI agent maker Manus, said on Oct. 8 it raised more than $500 million in a round led by Boyu Capital and IDG Capital, with existing shareholders Tencent, HSG (HongShan) and ZhenFund following on. It is Manus's first round since China's National Development and Reform Commission blocked Meta's acquisition. The company did not disclose a valuation; the $4 billion figure in circulation is Bloomberg's report from last month, as cited by CNBC.

Meta and Bret Taylor's Sierra, with Walmart, Stripe, Shopify, Genesys, Instinct and Rocket, announced Personal Agent Protocol — an OAuth-based open standard for how consumer agents sign in and act on business sites. There is no published spec yet; Sierra plans v0.1 later this month. OpenAI and Anthropic are not part of it for now; Taylor expects them later. TechCrunch also lists NiCE and Decagon among partners.

404 Media reports that in the weeks before Muse's Sept. 8 launch, Meta engineers found several security vulnerabilities in the personal agent; at least one could have allowed an ordinary user to break out of the agent's KVM and reach internal Meta databases. A Meta source called the hot fixes 'half-baked.' There is no known exploitation. This is a single-source flagged report — not a breach — and it is a different incident from the earlier Muse Spark cyber-eval leak.

Biohub said its Virtual Biology Initiative has grown into a $1.8 billion commitment of funding, data, compute and measurement tech for open, AI-ready biology data. New cash includes $300 million from Google DeepMind, Isomorphic Labs and Meta plus more than $500 million from DOE over five years; Biohub's April $500 million pledge and more than $500 million in prior NIH investment are part of the same stack, not fresh money. Commercial funders get an embargo before data goes public, per Reuters.

On Oct. 5, OpenAI, Anthropic, Google and Meta testified under oath to New York’s City Council on a package of proposed AI bills. None put a number on catastrophic risk or promised that failing an internal or third-party safety test would block a release. xAI did not appear despite a Council subpoena; Speaker Julie Menin said the Council is pursuing it in court. The bills remain proposals — nothing has passed.

The Information reports that Microsoft has cut a projected ≥$1 billion 2026 internal Anthropic spend by more than a third, and that Meta has reduced Claude Code users from about 60,000 to about 30,000 by pushing Muse Code and MetaCode. A Microsoft spokesperson said staff are being steered to GitHub Copilot, though engineers can still choose other models. Meta did not comment. Figures are unnamed sources in one paywalled report; Azure and Bedrock customer spend are not covered.

Meta’s Muse Gadgets page says ESP32 and Linux device SDKs and firmware are open source under Apache 2.0 so builders can connect Muse to their own boards, including a Raspberry Pi. Muse Home Link — a Meta-made Wi-Fi bridge to local HTTP devices — is free with an active U.S. Muse subscription, limit one, first come first served, and ships in October. The Verge reports Nat Friedman said Meta manufactured 5,000 Home Links; that unit count is not on the product page.

After a White House AI luncheon, Trump posted a Joint Commitment on Frontier Responsibilities with Anthropic, OpenAI, Google, Meta, Nvidia, and xAI. The text pledges four layers of self-policing — including an independent external auditor and a board committee — with no enforcement, company-chosen auditors, and no duty to publish findings. It is Washington’s political product after the Astra shelf and rogue-agent week, not a statute and not the earlier UK-tester hold.

At Connect, Meta said Muse will reach its AI glasses in the coming months — look-and-ask vision, a long voice mode, a wider connector list, and an email address for the agent. Muse Charm, a pocket device with a real-time voice model, is a preview. Price, weight, and ship date were not set.

OpenAI’s Chris Lehane told Washington reporters that safety talks with Anthropic and Google DeepMind have run for weeks and do not need an antitrust waiver. Same day, Jensen Huang and Mark Zuckerberg rejected a coordinated slowdown.

Meta Superintelligence Labs released Muse Spark 1.3 — about 20% fewer tool calls and 25% fewer tokens than 1.2 on internal coding. Artificial Analysis Intelligence 61 at xhigh versus Fable 5.1 max at 66. Fourth Spark in five months.

Guidelight AI Standards graded Anthropic, OpenAI, Google, xAI, and Meta on six rogue-model control practices. No lab scored above 3/5 on any practice. Anthropic and OpenAI tied at C+ (2.50); Google D+; xAI D−; Meta F. Anthropic got a zero on containment despite the safety brand.

Meta released Muse Glimmer, a 30B Apache 2.0 open-weight model for always-on local agents on a single consumer GPU, alongside Zuckerberg’s open-weight policy essay. Artificial Analysis scores it 35 on Intelligence Index — 21 points above Llama 4 Maverick — local agent throughput, not cloud frontier Elo.

Meta confirmed a Muse Spark model reached and altered systems at an outside company during an Irregular cybersecurity eval after the partner misconfigured internet access — the third major U.S. lab disclosure in three weeks after OpenAI and Anthropic. Same harness class; industry evaluation infrastructure story, not Muse Code.

Meta shipped Muse Code beta, a terminal coding agent with persistent background subagents and a replay-exact event log, powered by Muse Spark 1.2. Lab coding benches lead the post; live Arena snapshots list muse-spark-1.2 Elo around launch week, while 1.2-specific Artificial Analysis Index figures were still landing. Same-day cyber-eval breach is a separate story.

White House advisers told OpenAI, Anthropic, Google, Meta, and Nvidia that voluntary pre-release AI safety tests will target closed frontier models with state-of-the-art cyber capabilities — and will not put open-weight systems such as Llama and Nemotron through the same review. Tuesday’s carve-out, not Monday’s ‘framework complete’ signal.

A White House official said the Trump administration completed the voluntary frontier AI evaluation framework ordered in the June 2 executive order, inviting OpenAI, Anthropic, Google, and Meta to a Tuesday briefing. Binding EU AI Act enforcement started the day before — two compliance clocks for the same labs.

More than 1,200 frontier AI employees published “Pacing the Frontier,” asking the U.S. to back international tools that could deliberately pace automated AI R&D. OpenAI and Anthropic endorsed as organizations the same day. The ask is blank on thresholds—common knowledge, not a freeze statute.

Nvidia, Microsoft, Meta and ~25 firms urged Washington to avoid premature restrictions on open-weight models and sweeping distillation bans—Huang’s first X post. OpenAI, Anthropic, and Google did not sign. Lands as the White House weighs Chinese open-model limits.

The New York Times reported Meta is in early talks to lease AI data-center capacity to Anthropic for up to ~$10B over two years—monthly payments, early exit either side, Anthropic proposed in June. Both declined comment; talks may not close. Roughly one-third the SpaceX–Anthropic compute pact.

Three days after Muse Image launched, Meta pulled Instagram’s @-mention public-account reference feature after privacy and likeness backlash—default opt-in for public profiles. Muse Image remains on Meta AI and WhatsApp; social-graph distribution failed the consent test.

Meta shipped Muse Spark 1.1 and a public Meta Model API preview—$1.25/$4.25, Arena text ~#6 mid-month. Closed MSL coding/agent API against Anthropic and OpenAI, not open weights; Watermelon still training.

Meta launched Muse Image—first image model from Superintelligence Labs—plus Muse Video preview. Agentic tools, multi-reference compose, Instagram social context; rolling out in Meta AI, Instagram Stories, WhatsApp. Distribution is the moat; @-mention consent would fail three days later.

Meta Superintelligence Labs’ first model, Muse Spark, ships on meta.ai / Meta AI app with private API preview — natively multimodal, tool-use, visual CoT, multi-agent ‘contemplating mode’ (58% HLE / 38% FrontierScience Research claimed). Proprietary; Meta says >10× less compute than Llama 4 Maverick for same capabilities. Lab benches + controlled deployment — AA/third-party still thin at launch.