This is an earlier edition. Today's AI Daily →

DailyAIThu, 12 Nov 2026Published 09:45 UTC
AI for scientific discovery and biotech drug development
OHMEGA / MONSTERAI Daily Thu, 12 Nov 2026AI-generated image — not a photograph

DeepMind released AlphaGenome 2 and verified Signal's code as Fable 5.2 held six benchmark records and Gemini 4 Carbon claims stayed unverified

The day in 30 seconds

All situations side by side · state, direction, Pulses, coverage
State · move in 24hΔ 24h
  1. AI for scientific discoveryGoogle DeepMind on Nov 11 released AlphaGenome 2, a cell-type-conditioned DNA model that nearly doubled causal-variant recall on an immune-cell benchmark… rising— rising
  2. Compute, chipsAs of Nov 12, nothing has changed… steady
    model-assessed
    ◆ 0 falling
  3. Gemini 4 Carbon frontier pushThe picture since Nov 10 added Google's Gemini 4 Carbon rollout and Gemini Robotics 2.5 pilot claims… steady
    model-assessed
    ◆ 0 falling
  4. AI safety governance, evaluationPattern held since Nov 10: steady, gradual escalation around lab kill-switch controls… steady
    model-assessed
    ◆ 0 steady
  5. AI-enabled hackingAI-enabled hacking risk remains elevated as of Nov 12, unchanged since Nov 8 steady
    model-assessed
    ◆ 0 steady
The day, in more detail (5)
The day
  • Google DeepMind released AlphaGenome 2 on Nov 11 that nearly doubled causal-variant recall on an immune-cell benchmark…
  • As of Nov 12, independent leaderboards measured Claude Fable 5.2 at new records on six benchmarks including 1,836 on LMArena web development…
  • Google reportedly rolled out Gemini 4 Carbon with an agent swarm feature on Nov 10 and claimed Gemini Robotics 2.5 hit 96% sorting in a Mercedes-Benz pilot…
  • AI-enabled hacking risk stayed elevated as of Nov 12 with hundreds of exposed Artex agents still publicly reachable…

State of play · not in today's sections

AI for scientific discovery and biotech drug development · situation → , gradual pace

DeepMind's AlphaGenome 2 maps disease variants by cell type as AI lab benchmarks advance

Comparison AlphaGenome 2 nearly doubled causal-variant detection and directional-effect prediction, while GPT-6 Astra sharply outscored its predecessor on raw wet-lab data.. New benchmarks show AI models improving markedly at interpreting genetic variants and raw lab artifacts.deepmind.google ↗theverge.com ↗openai.com ↗as of 2026-11-12AI-assisted analytic, built only from real cited or sourced data

The rundown

AI for scientific discovery is moving from helping analyze data to helping design and run biology experiments itself. Anthropic, Google DeepMind and OpenAI are building agents that can search vast biological sequences, do chemistry and handle lab steps, working in lower-risk labs without human pathogens. The shift matters because when an AI can act in the lab, its capabilities translate directly into physical biological outcomes.

Google DeepMind on Nov 11 released AlphaGenome 2, a cell-type-conditioned DNA model that nearly doubled causal-variant recall on an immune-cell benchmark, and on Nov 10 showed AlphaProof Code verifying Signal's core in four weeks with over six million lines of machine-checked proof. OpenAI on Nov 6 set a new wet-lab baseline with BioBasicsBench, where GPT-6 Astra reached 66.7% but still struggled on images and controls. No new wet-lab synthesis or mouse seizure outcomes have been reported since Nov 4, so autonomy claims remain ahead of biological validation.

New · researched reported

  • Escalated: From agentic cheminformatics and a single autonomous mouse injection to broader platform releases…
  • New: First quantitative wet-lab benchmark (BioBasicsBench) and first cell-conditioned genome model (AlphaGenome 2) with head-to-head gains reported
  • 11 Novdeepmind.google ↗
  • 10 Novdeepmind.google ↗
  • techcrunch.com ↗
  • 6 Novopenai.com ↗

Nothing changed since yesterday: No new synthesis/assay results for proposed macrocycles or seizure-frequency data from the treated mouse; biological validation still pending as flagged on Nov 4.

Watch for

  • Independent use of AlphaGenome 2 API to replicate IBD ETS2 and other variant-to-function claims in new cell types
  • BioBasicsBench follow-ups: error analysis on image/control tasks and any model updates closing the gap
  • Wet-lab synthesis and assay results for DeepMind's proposed macrocycles and antimicrobial peptide candidate
  • Seizure-frequency data and any additional cohorts from the Claude autonomous mouse brain injection

Outlook AlphaGenome 2 is likely to see rapid academic uptake via the non-commercial API, but its higher compute cost will likely limit broad screening until efficiency improves. BioBasicsBench scores suggest incremental gains in wet-lab reasoning are likely, yet image and control failures make fully autonomous wet-lab loops unlikely in the near term. PSI-style theory wins are likely to continue as frontier models are tested, but translation to lab-validated biology remains the harder, slower step.

Compute, chips and supply: export controls and China's compute gap · situation → , flat pace

Chip export-control standoff unchanged as Nvidia probe and House scrutiny remain pending

The rundown

Since October 2022 the United States has tightened export controls on advanced computing and chipmaking technology for China, including limits on advanced memory and chipmaking equipment, while adding many Chinese firms to its trade blacklist. China is trying to build domestic alternatives and still buys chipmaking tools, while the AI chip supply chain runs from Nvidia designs through TSMC manufacturing and packaging in Taiwan. The contest matters because who controls advanced chips shapes AI leadership and broader economic and security competition.

As of Nov 12, nothing has changed since Nov 11 — and since Nov 9 — in the US chip export-control and China compute-gap standoff. Nvidia's independent board probe into alleged Chinese intelligence targeting of CEO Jensen Huang and House scrutiny remain pending with no new findings, BIS approvals, import clearances or seizure decisions reported. Confidence is high on the stasis; underlying spy-recruitment and chip-pile claims remain press and intelligence reporting, not charged or confirmed.

On the record

  • Jensen Huang · Chief Executive, Nvidia · 10 Nov

    “I support the investigation and will cooperate fully,”

    Commitment · conciliatory toward United States, Nvidia as reported by Deutsche Welle ↗

  • United StatesDonald Trump · President of the United States · 10 Nov

    “I just know Jensen very well. I just think he’s a very honourable man,”

    Reassurance · conciliatory toward Jensen Huang, Nvidia as reported by South China Morning Post ↗

  • Nvidia board · Board of Directors, Nvidia · 8 Nov

    paraphrase The board currently has no reason to believe that US national security or Nvidia's confidential information had been compromised, but will examine all available information in its independent investigation

    Reassurance · about United States, China as reported by Anadolu Agency ↗

All on the record (10) · Compute, chips and supply: export controls and China's compute gap, across the day's updates
  • Jensen Huang · Chief Executive, Nvidia · 10 Nov

    “I support the investigation and will cooperate fully,”

    Commitment · conciliatory toward United States, Nvidia as reported by Deutsche Welle ↗

  • United StatesDonald Trump · President of the United States · 10 Nov

    “I just know Jensen very well. I just think he’s a very honourable man,”

    Reassurance · conciliatory toward Jensen Huang, Nvidia as reported by South China Morning Post ↗

  • Nvidia board · Board of Directors, Nvidia · 8 Nov

    paraphrase The board currently has no reason to believe that US national security or Nvidia's confidential information had been compromised, but will examine all available information in its independent investigation

    Reassurance · about United States, China as reported by Anadolu Agency ↗

  • Jensen Huang · Chief Executive, Nvidia · 10 Nov

    “I am committed to protecting NVIDIA and supporting America's national security.”

    Reassurance · conciliatory toward United States, Nvidia as reported by Deutsche Welle ↗

  • United StatesDonald Trump · President, United States · 10 Nov

    “He said he had absolutely no idea that the person in question was a spy,”

    Reassurance · constructive toward United States, Jensen Huang as reported by Deutsche Welle ↗

  • Nvidia · company, Nvidia · 9 Nov

    “target of a foreign intelligence effort”

    Policy · about Nvidia, China as reported by Deutsche Welle ↗

  • United StatesDonald Trump · President, United States · 2 Nov

    paraphrase He is ready to meet with his Russian and Chinese counterparts in November specifically to discuss arms control and denuclearization.

    Offer · constructive toward Russia, China, United States as reported by Kyiv Independent ↗

  • United StatesDonald Trump · President of the United States · 10 Nov

    “I asked him that question. Did you have any idea that she was a spy? He said I had no idea that she was a spy, and I really believe him,”

    Reassurance · about China, United States as reported by South China Morning Post ↗

  • United StatesDonald Trump · President of the United States · 10 Nov

    “No, I didn’t hear about it then.”

    Denial · about China as reported by South China Morning Post ↗

  • United StatesDonald Trump · President of the United States · 10 Nov

    paraphrase Trump said he did not think he would raise the Nvidia espionage case with Chinese President Xi Jinping at their expected meeting next week, but suggested Xi might raise it.

    Remark · about China, Xi Jinping as reported by South China Morning Post ↗

A statement shows what was said, not that it is true.

New

    Nothing changed since yesterday: No new findings, approvals or enforcement actions since Nov 11; board probe and House scrutiny still pending

    Watch for

    • Nov 11 deadline for Huang to accept or propose alternative date for House closed-door briefing — any response or rescheduling
    • Nov 18 joint House briefing on Huang interactions and Nvidia security protocols
    • Nov 17 Nvidia Q3 earnings for H200 China revenue and any board probe update
    • Any Hong Kong or South Korea customs/seizure data on HBM smuggling and any BIS vetting changes for H200 handlers

    Outlook Stasis is likely to hold until the next forcing events around Nov 17-18. A House briefing response, BIS vetting update, or customs/seizure data would likely shift the picture; absent those, the strained but steady posture is likely to continue.

    Economies involved · size, trade, energy, arms · annual estimates

    Gemini 4 Carbon frontier push and challengers · situation → , gradual pace

    Unverified report says Google launched Gemini Decision 1 to score 1,000 questions in one pass

    The rundown

    The frontier-model race pits Google's Gemini family, including Gemini 4 Carbon and Gemini Robotics, against Anthropic's Claude models, Meta's Muse, ByteDance's Seedream and others, each trying to claim the most intelligent and efficient model on benchmarks, cost and speed. Leadership matters because the models are increasingly used to build the next generation of models. Anthropic describes this with an automation scale and defines recursive self-improvement as an AI fully designing its successor, a threshold it says has not been reached.

    The picture since Nov 10 added Google's Gemini 4 Carbon rollout and Gemini Robotics 2.5 pilot claims, plus a single unverified report of Gemini Decision 1 scoring 1,000 questions in one pass. We are confident Fable 5.2's 5-point lead on Artificial Analysis Intelligence Index v5 (65 vs 60) and its sweep of Ohmega-tracked benchmarks still stands; Decision 1 remains unconfirmed with no lab post, pricing, or independent Artificial Analysis/LMArena/Terminal-Bench replication, so leadership is unchanged pending verification.

    New · researched reported

    • New: Single unverified ai_decoder report Nov 11 of Gemini Decision 1 scoring 1,000 questions in one pass — no lab post or independent replication to verify
    • New: Gemini 4 Carbon now available with agent swarm (Nov 10 DeepMind blog) and Gemini Robotics 2.5 dexterity claims including 96% Mercedes-Benz pilot…
    • 11 Novthe-decoder.com ↗
    • 10 Novdeepmind.google ↗
    • deepmind.google ↗

    Nothing changed since yesterday: Fable 5.2's 5-point Artificial Analysis lead and Ohmega-tracked benchmark sweep unchanged; watched LMArena/Terminal-Bench replication and pricing/rollout after Carbon introductory period still pending

    Watch for

    • Google DeepMind lab post, model card or pricing for Gemini Decision 1 or Gemini 4 Carbon confirming Decision 1 exists and its cost
    • Independent Artificial Analysis Intelligence Index v5, LMArena or Terminal-Bench replication of Fable 5.2 vs Carbon and any Decision 1 batched-scoring verification
    • Anthropic Responsible Scaling Policy update or METR follow-up on RSI threshold and acceleration

    Outlook Decision 1 is likely to stay unverified until Google publishes a lab post with pricing and model card and independent Artificial Analysis, LMArena or Terminal-Bench replication appears; without that, Fable 5.2's lead almost certainly stands. If Google confirms batched scoring with competitive cost and accuracy, the frontier could shift from single-answer benchmarks to decision-model throughput, but that is roughly even pending evidence. Coverage falling (6 vs 10 headlines) suggests a pause before next verified scores.

    AI safety governance, evaluation and catastrophic-risk regulation · situation → , gradual pace

    AI safety governance held steady with no new laws, pledges or enforcement actions since November 10

    The rundown

    Frontier AI is a race among OpenAI, Google DeepMind and Anthropic to lead in coding and enterprise work while setting rules for catastrophic risk. Anthropic has tried to measure the race by tracking how much of its own research and code is done by its AI and by publishing tiered tests for cyber, biological, sabotage and loss-of-control risks. The central tension is recursive self-improvement — an AI that can fully design its successor — and whether evaluations can keep up if capabilities jump faster than safeguards.

    Pattern held since Nov 10: steady, gradual escalation around lab kill-switch controls, DOJ and EU oversight pressure, and bioweapon warnings continues, with no new law, pledge, RSI decision or enforcement action evidenced in the last three days. Coverage was steady at 5 headlines vs 4 before, reflecting continued attention rather than a new shift.

    On the record

    • United StatesDonald Trump · US President, United States · 3 Nov

      “can buy whatever they want”

      Offer · constructive toward Israel, Europe as reported by CGTN ↗

    New

      Nothing changed since yesterday: No new law, pledge, RSI decision or enforcement action evidenced since Nov 10; prior picture stands: OpenAI Control Room kill-switch deployed, DOJ-led enforcement rhetoric on record, EU Commission information requests pending, and bioweapon/sabotage warnings elevated.

      Watch for

      • Anthropic update to Responsible Scaling Policy or formal RSI/2x decision, and any less-redacted METR evidence or independent replication of 1.9x estimate
      • Independent replication of Control Room coverage (unmonitored traffic share, RL monitoring) and of GPT-6 Astra covert-underperformance fix
      • EU or company disclosure responding to Commission request: incident counts, EU impact, and changes to risk assessments/security practices

      Outlook Almost certain that oversight pressure around kill-switches, incident reporting and bioweapon risk stays elevated in the near term, given the New York RAISE Jan 1 effective date, Paris Call follow-on standards work, and pending EU information requests. A formal RSI determination, new enforcement action, or independent replication of METR/Control Room claims remains unlikely to appear without a lab or regulator disclosure, roughly even in the next week.

      Economies involved · size, trade, energy, arms · annual estimates

      AI-enabled hacking and agentic cyber offense · situation → , flat pace

      AI hacking risk stays elevated from exposed Artex agents and near-parity open models with no new mass exploitation confirmed

      Comparison Open-weight model reached near-parity on cyber-offense objectives at a fraction of frontier cost while exposed agent infrastructure remains widespread.. Frontier and open-weight models now score closely on cyber-offense tasks, but open models run far cheaper.vals.ai ↗censys.com ↗openai.com ↗as of 2026-11-12AI-assisted analytic, built only from real cited or sourced data

      The rundown

      AI labs are building models that can act as autonomous hacking agents, while browser and platform defenders try to protect hundreds of millions of users. Open-weight models whose weights are public can be run without provider controls and have narrowed the gap with closed frontier systems. The key leverage is the patch gap, when a fix is visible in code but not yet installed on most devices, letting an agent find and chain flaws to break out of the browser sandbox and steal data.

      AI-enabled hacking risk remains elevated as of Nov 12, unchanged since Nov 8. Hundreds of exposed Artex agents remain publicly reachable and open-weight models remain within single-digit points of frontier closed models at roughly one-tenth the cost, while defender hardening continued. No new mass patch-gap exploitation at scale was confirmed beyond previously linked Artex incidents.

      New · researched reported

      • 11 Novgov.uk ↗
      • 10 Novopenai.com ↗

      Nothing changed since yesterday: Pattern continued since Nov 8: elevated risk from ~800+ distinct exposed Artex agents and near-parity open models persists; defender hardening advanced (Cyber Defense Factories, Pall Mall guidelines) with no new mass patch-gap exploitation at scale confirmed

      Watch for

      • Censys or other scans showing Artex exposure count falling or evidence of exploitation of exposed consoles
      • Mozilla and Linux distro promotion of GPT-6 Cyber-found fixes to stable and time to reach most users
      • Evidence of GPT-6 Cyber or Volta-style LLM ranking used in wild exploitation or monetization beyond lab samples

      Outlook Risk likely stays elevated in the near term while exposed Artex consoles remain reachable and open models offer near-frontier hacking at low cost. Defender factory deployments and Pall Mall guidelines are unlikely to quickly shrink the exposed surface or patch-gap window. No new mass exploitation at scale is expected to be confirmed unless scans show exploitation of exposed consoles or patch-gap chaining in the wild.

      Economies involved · size, trade, energy, arms · annual estimates

      Across situations

      Also watching

      Quiet