The biggest AI news, October 15, 2026 | AI News - Dataspheres AI

The biggest AI news, October 15, 2026 AI NEWS The biggest AI news, October 15, 2026 October 15, 2026 Welcome to AI News. Today's rundown Today's rundown Me...

The biggest AI news, October 15, 2026 AI NEWS The biggest AI news, October 15, 2026 October 15, 2026 Welcome to AI News. Today's rundown Today's rundown Meta and Google ship frontier models two days apart AI money splits: record revenue, punished stocks Security: self-copying AI and a bugged boardroom Agents get a shopping cart The four segments of this edition, in order. Meta ships its first frontier models Meta ships its first frontier models — Muse Wonder and Muse Wonder Full, October 12 Muse Wonder and Muse Wonder Full, launched October 12 Reuters Meta launched Muse Wonder and Muse Wonder Full on Monday October 12, calling them its most capable models yet, frontier models for agents. Muse Wonder will power the Muse personal agent for subscribers; Muse Wonder Full is for Muse Code subscribers. Sources: Reuters via Yahoo Finance; Fortune Tech Oct 13. Meta lands two points behind #1 Meta lands two points behind #1 — Muse Wonder Full: 56, #4 overall — 2 points behind Claude Opus 5.5 — Meta is back in the frontier race Artificial Analysis Intelligence Index (best configuration per model): Claude Opus 5.5 58, Claude Sonnet 5.5 56, Muse Wonder Full 56 (rank 4 across configurations), GPT-6 Astra 53, Gemini 4 Argon 53. Source: artificialanalysis.ai/leaderboards/models, read October 15. Google opens Gemini 4 Argon to developers Google opens Gemini 4 Argon to developers — Generally available October 13 Generally available October 13 Google via Android Headlines Gemini 4 Argon went generally available on October 13 on the paid Gemini API tier, two weeks after a limited release to cyber defenders. Google pitches it for complex coding, heavy reasoning, cybersecurity and robotics; 87% on Embodied Reasoning. Sources: Android Headlines Oct 14; ComputingForGeeks Oct 13. Argon's launch price doubles on December 1 Argon's launch price doubles on December 1 — Matches GPT-6.1 Sol until November 30 — Then costs the same as Claude Opus 5.5 — No free tier Argon: $2 input / $10 output per million tokens until November 30, 2026; $4 / $20 from December 1. GPT-6.1 Sol lists at $2 / $10; Claude Opus 5.5 at $4 / $20. No free tier. Source: ComputingForGeeks, tested October 13 against the GA model ID. AI money splits: record revenue, punished stocks AI money splits: record revenue, punished stocks — Better models cost money to serve, and the market asked who pays CoreWeave shares fell 15.5% on October 14 CNBC Bridge from the model launches to the money that runs them. Source: CNBC Oct 14. Wednesday's damage: AI names fell far more than the market Wednesday's damage: AI names fell far more than the market — S&P 500 -1.11%: its worst day since July — CoreWeave fell 14 times as far — Trigger: a hot inflation report and Fed rate-hike bets October 14: S&P 500 -1.11%, Nasdaq -1.88% (worst day since July); semiconductor index -4.3% (about -12% for the week); Nvidia -3.2%; Marvell -8.7%; Oracle -8.8%; Dell -11.3% (CNN). CoreWeave shares -15.5% (CNBC). A hotter-than-expected inflation report bolstered bets on further Fed rate hikes; higher yields make the data-center buildout more expensive. Borrowing is the pressure point Borrowing is the pressure point 14.9% Yield on CoreWeave's 2031 bond, against its 9.75% coupon -10%+ SK hynix on October 13, as AI jitters hit Asia CoreWeave 9.75% bond due 2031 yielding about 14.9%; 9.625% 2032 bond at 14.8% (CNBC Oct 14). CEO memo quote via CNBC. SK hynix fell more than 10% and Samsung 7.4% on October 13 (Tech in Asia). Meanwhile, Anthropic's revenue keeps climbing Meanwhile, Anthropic's revenue keeps climbing $16.8bn Anthropic's latest quarterly revenue, up nearly 50% on the quarter before Anthropic told investors quarterly revenue rose nearly 50% to $16.8bn ahead of its IPO (Financial Times, October 14, headline). Anthropic: small AI models may soon copy themselves Anthropic: small AI models may soon copy themselves — From the money to the risk: two security wake-up calls Anthropic research, published October 12-13 Ars Technica Bridge to the security segment. Sources: Reuters syndications (Khaleej Times, The Star, Oct 13), Ars Technica Oct 14. Why it matters: there is no central off-switch Why it matters: there is no central off-switch Small models (under 30B parameters) completed each replication step, but not the whole chain yet A demo ran a model across 15 Android TV boxes costing about $55 each Anthropic's researchers: possibly months away Their ask: harden cheap connected devices now Anthropic researchers William Zhou and Daniel Freeman: small models may be months away from self-replication on commercial IoT devices. Small models under 30B parameters could complete the many steps but struggled to complete the whole sequence; a replicator would be hard to shut down because there is no central point to switch off. Demo: Gemma 4 26B-A4B across 15 Android TV boxes at about $55 each. Sources: Reuters via Khaleej Times; AI Magazine. An OpenAI engineer is arrested over a hidden recorder An OpenAI engineer i