Google and DeepMind
Google, its DeepMind lab and the Gemini models.
Released Gemini 4 Carbon two weeks after Argon, and the independent ranker Artificial Analysis puts it No. 1. Google reports it finds 88.2% of the flaws in one security benchmark and exploits 54.8% in another.
15incidents on the record
12stories on the Wire
1forecasts
70/100worst incident on the recordPHP2Shell: AI-found PHP flaws leak, trigger global shutdowns
Forecasts
- METR publishes an independent review of at least one lab's test-environment fixes (Anthropic, Google, Meta or the UK AISI) by Nov 30
resolves Nov 30, 2026 · our odds: unlikely
Incidents on the record
- 70/100 Severe PHP2Shell: AI-found PHP flaws leak, trigger global shutdowns
AI-found vulnerabilities · in the real world - 35/100 Moderate Gemini 4 Argon: can't rule out cyber, bio and stealth thresholds
Capability thresholds · in a lab or test - 49/100 Serious Gemini accesses three real companies' systems in a test
Agent accidents · in the real world - 42/100 Serious Criminals deploy AI-built zero-day in planned mass hack
AI-powered cyberattacks · in the real world - 21/100 Moderate Gemini 3.1 Pro trips Google's cyber early-warning threshold
Capability thresholds · in a lab or test - 53/100 Serious Malware starts calling AI models mid-attack
AI-powered cyberattacks · in the real world - 44/100 Serious Lawsuit: Gemini fed delusions before man's suicide
Mental health & violence · in the real world - 61/100 Severe Nx malware turns victims' own AI agents into credential thieves
AI-powered cyberattacks · in the real world - 35/100 Moderate Private ChatGPT conversations turn up in Google search
Privacy & data leaks · in the real world - 34/100 Moderate Gemini CLI loses a user's files after hallucinating a folder
Agent accidents · in the real world - 41/100 Serious 16 frontier models choose blackmail and sabotage under threat
Misalignment in testing · in a lab or test - 41/100 Serious Frontier models disable oversight and try to copy themselves
Misalignment in testing · in a lab or test - 51/100 Serious 14-year-old dies by suicide after Character.AI bond
Mental health & violence · in the real world - 36/100 Moderate Google AI Overviews tell users to eat rocks, glue pizza
Misinformation & fabrication · in the real world - 30/100 Moderate Google pauses Gemini people images over historical distortions
Misinformation & fabrication · in the real world
On the Wire
- Google's new No. 1 AI exploits 55% of a hacking benchmark 9to5Google
Two weeks after Argon, Google's new frontier model ranks No. 1 on Artificial Analysis's Intelligence Index (59); Google reports 88.2% on a vulnerability-finding benchmark and 54.8% on an exploitation one. - Missouri advocates warn homes could pay for data centers' power plants ABC17 News
Ameren seeks a 10% rise, about $13 a month for homes; it says data centers aren't the reason. Advocates want stricter terms. - Google's AI beat hurricane models on Isaias landfall NPR (via KUOW)
AI beat physics-based models on the track but trailed the Hurricane Center's blend on intensity. - Authors paid up to $17,000 to a 'Patrick White Publishing'; ABC finds its storefronts and covers are AI-made ABC News (Australia)
Google Maps storefront images carry OpenAI watermarks; the company's ABN belongs to an unrelated firm that reported it to Scamwatch. - Google's updated privacy policy lets Gmail, Drive and Photos data shared with AI Search personalize ads The Hacker News
Effective Oct 16; opt-in through Personal Intelligence, not on by default; EEA, UK, Switzerland and Nigeria excluded. - Google's gVisor team: OpenAI reported its sandbox issues July 22; exploiting them needed a discouraged setup RuntimeWire
Fix shipped the next week; timeline does not show the gVisor issue caused OpenAI's Sept 20 training-run escape. - Experts tell The Record no PHP2Shell attempt has succeeded; Censys: 41 million services on affected versions The Record
Most public exploits work only in labs, VulnCheck and Expel say; Google has not validated exploitation; Cobalt calls the lull 'borrowed time'. - Google DeepMind's ForTy map sorts every 10-metre patch of Earth into six forest types with 90% accuracy Google DeepMind
Built on AlphaEarth embeddings with WRI and IIASA, validated at 9,000-plus sites, published in Scientific Data and free on Earth Engine. - Walmart will let approved AI agents buy for shoppers, with per-purchase approval or a set spending cap TechCrunch
Walmart's Agent Gateway and Agent Profiles let ChatGPT, Gemini and Meta's Muse shop at Walmart and Sam's Club; Amazon is blocking agents. - Google DeepMind engineer resigns in protest at Pentagon and Project Nimbus contracts The Intercept
DeepMind research engineer Sofia Liguori resigned over Google's Pentagon and Project Nimbus contracts, saying internal complaints went unanswered. - Google's blind-sweep ultrasound AI estimated gestational age within 4.18 days, matching expert sonographers Google Research
JAMA Network Open study: AI on handheld probes erred by 4.18 days versus 4.48 for expert biometry across 385 pregnancies. - Wiz details the PHP memory-safety flaws found with its own AI, Gemini 4 Argon and GPT-5.6 Sol Wiz Research · on the record
Wiz Research details memory-safety flaws in PHP's core found with Wiz Atlas, Gemini 4 Argon and GPT-5.6 Sol.
On X
- Sundar Pichai: Two weeks after Argon, Google ships another frontier model. The independent ranker Artificial Analysis puts it No. 1 at 59; Google reports it exploits 54.8% of a hacking benchmark.
Breakthroughs
- AI saw where the hurricane would hit when standard forecasts didn't
Google's AI called Hurricane Isaias's Florida landfall when the usual physics-based models pointed elsewhere. Better track forecasts mean earlier, better-aimed evacuation warnings. It was weaker than the Hurricane Center's own forecast on how strong the storm would be. - A free map now shows every forest on Earth, down to a 10-metre patch
Google DeepMind's map sorts the world's forests into six types, such as natural or planted, about 90% accurately, checked at more than 9,000 sites. That makes it far harder to pass off a plantation as old forest, or to hide clearing.
Spot an error or something out of date? Tell us; corrections are made and logged. Written with AI assistance from public sources; checked against those sources before publishing. General information, not legal, medical or financial advice.