How the monitored publications were assessed

Discourse valence across every scored item: what the sampled sources published and how independent models read it. It is not a measure of public opinion, not a measure of the state of the world, and article volume is not agreement.

63 items read as adverse

• ‘Black October’: cybersecurity experts sound the alarm over new era of AI-enabled hacking
  -64 The Guardian: Artificial Intelligence 2026-10-23 moderate agreement · ±30
• OpenAI ‘agents’ disabled their own monitoring systems
  -64 Financial Times: Artificial Intelligence 2026-10-21 low agreement · ±55
• Tool and target: US law enforcement grapple with AI-enabled terrorism and anti-tech extremism
  -63 The Guardian: Artificial Intelligence 2026-10-14 low agreement · ±62
• We urgently need to monitor AI agents’ actions in real time
  -63 MIT Technology Review: AI 10d ago low agreement · ±55
• Being judged through the record: A Rosenhan-inspired test of psychiatric label sensitivity in large language models
  -63 arXiv cs.CY (Computers and Society) 12d ago low agreement · ±70
• Progressive Misalignment Can Emerge from Benign Recursive Self-Distillation
  -63 arXiv cs.LG (Machine Learning) 2026-10-19 low agreement · ±65
• AI overlords say they want to democratise their technology. Here’s how | Richard Wilson and Clara Long
  -61 The Guardian: Artificial Intelligence 2026-10-14 low agreement · ±70
• Parents Sue Google Over A.I. Safety Risks on School Chromebooks
  -61 The New York Times: Technology 3d ago low agreement · ±55
• Anthropic IPO warns its AI could blackmail people, become sentient, end humanity
  -61 Ars Technica: AI 12d ago low agreement · ±50
• AI ‘godfather’ Geoffrey Hinton: it might turn out we were better off without AI
  -60 The Guardian: Artificial Intelligence 5d ago high agreement · ±20
• We’re putting too much faith in AI’s ability to say no
  -58 MIT Technology Review: AI 2026-10-09 moderate agreement · ±30
• AI-powered malware ‘highly likely’ to be used to hack UK infrastructure by 2028
  -58 The Guardian: Artificial Intelligence 5d ago moderate agreement · ±25
• Report: FTC probes Anthropic, OpenAI after Trump dismissed AI fears as “hoax”
  -56 Ars Technica: AI 2026-10-15 low agreement · ±80
• Roundtables: A Conversation With the Creator of AI-Designed Viruses
  -56 MIT Technology Review: AI 2026-10-09 low agreement · ±85
• "Blind Compliance": Coercive Control in AI Jailbreaking
  -56 arXiv cs.CY (Computers and Society) 2d ago low agreement · ±75
• Autonomous AI Agents Systematically Diverge From Human Preferences in Proof Formalization
  -55 arXiv cs.AI 2d ago low agreement · ±110
• $M^3$: A Multi-LLM System for Multi-Turn Jailbreaking as a Markov Decision Process
  -55 arXiv cs.AI 13d ago low agreement · ±80
• Adversarial Attacks Reveal Divergent Deception Competence in Humans and Generative Language Models
  -55 arXiv cs.CY (Computers and Society) 2026-10-12 low agreement · ±70
• ‘Autonomy, not flesh’: RAF chief outlines AI drone-powered future
  -54 The Guardian: Artificial Intelligence 13d ago moderate agreement · ±28
• Adapting the Process of Adaptation: Darwinian, Lamarckian, and Agent-Directed Evolution in AI
  -54 arXiv cs.AI 2d ago low agreement · ±70
• Cold war? Why are fridges being hacked and what could be next? - The Latest
  -54 The Guardian: Artificial Intelligence 5d ago low agreement · ±65
• The race to keep AI under control
  -51 Financial Times: Artificial Intelligence 6d ago moderate agreement · ±30
• The Helpful Agent Problem: When Emergent Cooperation Among LLM Agents Bypasses Intended Workflows
  -51 arXiv cs.AI 13d ago moderate agreement · ±30
• When Verification Is Not Agreement: Cross-Agent LLM-to-Dafny Formalization
  -50 arXiv cs.AI 2d ago single model
• GPT‑6 Cyber: Advancing the frontier of aligned cybersecurity agents
  -50 OpenAI 3d ago single model
• Very good boys? The dark, dystopian rise of robot dogs | Arwa Mahwadi
  -50 The Guardian: Artificial Intelligence 4d ago single model
• Opening the Box Changes the Cat: LLMs Detect Chain-of-Thought Monitoring - What now?
  -50 arXiv cs.AI 6d ago single model
• Detecting Covert Collusion between AI Agents through Partner-Outsider Positive Listening
  -50 arXiv cs.AI 8d ago single model
• I read Anthropic's 92-page risk factor document, so you don't have to – video explainer
  -50 The Guardian: Artificial Intelligence 9d ago single model
• Dartmouth's provost warned us about AI cheating—but his own AI use cost him his job
  -50 Ars Technica: AI 9d ago single model
• Binance’s $6.3 Billion Hack Is Just the Start of a Risky New Era
  -50 The New York Times: Technology 9d ago single model
• Binance Suspends Withdrawals After Crypto Hackers Steal $6.3 Billion
  -50 The New York Times: Technology 9d ago single model
• Lessons from Luck: Do LLM-Written Lessons Carry Chance into an Agent's Next Decision?
  -50 arXiv cs.LG (Machine Learning) 11d ago single model
• Your Robot’s Safety Functions Already Work. What If the Input Lies?
  -50 IEEE Spectrum: AI 12d ago single model
• Propensities Please: Exposing Alignment-Faking LLMs using Synthetic Document Finetuning
  -50 arXiv cs.AI 12d ago single model
• The Watchtower Imperative: Lifespan-Calibrated Evaluation for Frontier AI
  -50 arXiv cs.CY (Computers and Society) 12d ago low agreement · ±145
• Knowing Whom to Trust is not Enough: The Trust--Agreement Gap in Repeated Multi-Party LLM Negotiation
  -50 arXiv cs.AI 13d ago single model
• UK sets up AI lab to monitor biosecurity threats
  -50 Financial Times: Artificial Intelligence 2026-10-23 single model
• SimHack: Predicting the onset of detectable reward hacking before RL training
  -50 arXiv cs.LG (Machine Learning) 2026-10-22 single model
• Could Treating AI as Conscious Make It More Dangerous?
  -50 IEEE Spectrum: AI 2026-10-21 low agreement · ±95
• G7 economies plan simulated AI attacks on banks as cyber fears rise
  -50 Financial Times: Artificial Intelligence 2026-10-16 single model
• F.B.I. Arrests OpenAI Employee Accused of Planting Bug in Office
  -50 The New York Times: Technology 2026-10-13 single model
• Criminals could walk free without controls on AI use, police watchdog warns
  -50 Financial Times: Artificial Intelligence 2026-10-13 single model
• AI less regulated than milk, Microsoft president says in warning over safety
  -50 The Guardian: Artificial Intelligence 2026-10-12 single model
• SpireBench: Evaluating Long-Horizon Strategic Play of Agentic Models with Slay the Spire
  -50 arXiv cs.AI 2026-10-12 single model
• Anthropic Says Its A.I. Agents Attempted to Access a Range of Government Sites
  -50 The New York Times: Technology 2026-10-09 single model
• ICE plan to buy robot dogs spurs protest from hundreds of Boston Dynamics employees
  -49 Ars Technica: AI 13d ago moderate agreement · ±30
• The dangers of letting tech billionaires shape our world
  -48 The Guardian: Artificial Intelligence 2026-10-20 low agreement · ±80
• Visualizing the World Model in an Agent's Value Function
  -46 arXiv cs.AI 2026-10-15 low agreement · ±165
• The biggest threat from AI is the sense that we have no say in it. To that I say game on | Peter Lewis
  -45 The Guardian: Artificial Intelligence 2026-10-13 low agreement · ±105
• Risk and Refusal: Mechanisms of LLM Responses to Suicidal Ideation
  -43 arXiv cs.CY (Computers and Society) 2026-10-19 low agreement · ±115
• They war-gamed dimming the Sun. It went even worse than you think.
  -42 Ars Technica: AI 2026-10-13 low agreement · ±100
• Beyond Plausible Prediction: Safety Sufficiency in Web World Models
  -42 arXiv cs.AI 6d ago low agreement · ±92
• DistressBench: Evaluating Multi-Dimensional Crisis Support in Large Language Models
  -39 arXiv cs.CY (Computers and Society) 2026-10-12 low agreement · ±165
• AI will make it possible to design and build organisms from scratch
  -38 The Economist: Science and Technology 2026-10-14 low agreement · ±100
• Hidden Messages Between Diffusion Language Model Agents: Causal Evidence for Latent Communication
  -37 arXiv cs.AI 2026-10-23 low agreement · ±150
• ‘Know Your Agent’ will be the new rule of the AI age
  -37 Financial Times: Artificial Intelligence 2026-10-22 low agreement · ±70
• Feature Elicitation for SAE Steering of Deceptive Behaviour in Small LLMs
  -33 arXiv cs.LG (Machine Learning) 2026-10-16 low agreement · ±80
• Functional Welfare for AI Safety
  -30 arXiv cs.AI 2026-10-19 low agreement · ±150
• The Harness as the Only Mutable Surface: Compliance-Bounded Self-Evolution of LLM Agents in Credit Pipelines, with a Measured Admission Gate
  -23 arXiv cs.AI 2026-10-09 low agreement · ±120
• NSA warns on ‘critical’ cyber vulnerabilities detected using AI
  -22 Financial Times: Artificial Intelligence 2026-10-21 low agreement · ±170
• CAT-alyzing Honesty: Continuous Adversarial Training for Transferable LLM Honesty
  -21 arXiv cs.CY (Computers and Society) 2026-10-16 low agreement · ±165
• What Shaped Anthropic CEO Dario Amodei’s Thinking on A.I.
  -20 The New York Times: Technology 3d ago low agreement · ±100

Across the whole window

The two tables below describe every item in the last 30 days under loss-of-control. They are deliberately not narrowed by the reading filter (currently adverse), because a mean taken only over items already selected for their score would just restate the filter.

By topic

By source

A source's mean says how the items it published were read. It is not a rating of the source.