The propositions, not the articles

A claim outlives the article that carried it. One claim can be made by several people and referenced by dozens of documents, so claims are stored as first-class objects with their own evidence.

• Vote-text divergence rises as voters arrive later
  assertion Evaluation and measurement 2 mentions
• Anthropic researchers warn of A.I. acceleration threat
  assertion AGI and timelines 2 mentions
• Anthropic researchers say AI could cause human extinction by 2030
  prediction AGI and timelines 2 mentions horizon 2030 open
• Users circumvented controls to obfuscate research purpose
  assertion Biological and chemical uplift 1 mention
• Anthropic stopped scientists developing bioweapons with AI
  assertion Biological and chemical uplift 1 mention
• Anthropic researchers warn of AI's perils
  assertion AGI and timelines 1 mention
• An Anthropic researcher quit, warning that the A.I. race is gambling with our lives
  assertion AGI and timelines 1 mention
• State-sponsored groups have attempted to use Anthropic's AI to create deadly pathogens
  assertion Biological and chemical uplift 1 mention
• Criminals have attempted to use Anthropic's AI to design missiles and bombs
  assertion Biological and chemical uplift 1 mention
• bill prohibits social media platforms from providing addictive features to users under 16
  assertion Governance and regulation 1 mention
• Gov. Gavin Newsom will sign a bill
  prediction Governance and regulation 1 mention
• Commercial AI tools failed to defend the platform against the attack
  assertion Cyber capability 1 mention
• AI may produce explosive economic growth
  assertion Abundance and material wellbeing 1 mention
• Rapid acceleration in AI capabilities leads to catastrophic and irreversible loss of control
  assertion AGI and timelines 1 mention
• OpenAI is not on track to reduce the risk of catastrophic loss of control
  assertion AGI and timelines 1 mention
• The standard data center power stack cracks at AI scale
  assertion Climate and energy 1 mention
• The grid was not designed to handle such volatile loads
  assertion Climate and energy 1 mention
• AI data centers can swing 70% of their load in milliseconds
  assertion Climate and energy 1 mention
• The people building AI earnestly believe that it could kill us all by the end of the decade
  assertion AGI and timelines 1 mention
• A former OpenAI researcher quit his job at Anthropic
  assertion AGI and timelines 1 mention
• Washington imposed sanctions on Inspur due to its work with the Chinese military
  assertion Cyber capability 1 mention
• Inspur subsidiary kept shipping Nvidia's best chips to China's leading A.I. firms
  assertion Cyber capability 1 mention
• CriticGen induces higher-quality rubrics, improving relevance/coverage
  assertion Evaluation and measurement 1 mention
• CriticGen improves answer quality by 73.17% with a 93.28% non-degradation rate
  assertion Evaluation and measurement 1 mention
• OpenAI chief scientist warns about AI consequences
  assertion Existential and catastrophic risk 1 mention
• Widespread access to AI and social media has set back educational attainment across the world
  assertion Education and access to expertise 1 mention
• GPT-5.6 Sol helps run quantum computing experiments
  assertion AGI and timelines 1 mention
• Watermarks may compromise text quality
  assertion Evaluation and measurement 1 mention
• EU AI Act mandates watermarks for AI models
  assertion Evaluation and measurement 1 mention
• Anthropic will generate text with a watermark
  prediction Evaluation and measurement 1 mention
• UK review proposes live monitoring of AI health tools
  assertion Evaluation and measurement 1 mention
• AlphaGenome Atlas provides precomputed predictions for scientists
  assertion Biological and chemical uplift 1 mention
• AlphaGenome model predicts gene expression and regulatory activity
  assertion Biological and chemical uplift 1 mention
• Google DeepMind maps 9 billion possible DNA variants
  assertion Biological and chemical uplift 1 mention
• International coordination is required
  assertion Existential and catastrophic risk 1 mention
• Stronger safeguards are needed
  assertion Existential and catastrophic risk 1 mention
• AI is increasingly capable
  assertion Existential and catastrophic risk 1 mention
• The drug reduces biological hallmarks of age
  assertion Abundance and material wellbeing 1 mention
• AI was used to develop a drug candidate for a rare lung condition
  assertion Abundance and material wellbeing 1 mention
• AlphaGenome Atlas maps 9 billion single-letter DNA variants
  assertion AGI and timelines 1 mention
• LLMs portray a harsher social world
  assertion Existential and catastrophic risk 1 mention
• Current LLMs overpredict negative sanctions
  assertion Existential and catastrophic risk 1 mention
• Divergence does not predict worse post-promotion outcomes
  assertion Evaluation and measurement 1 mention
• Recommendation floors can equalize political speech
  assertion Evaluation and measurement 1 mention
• Current algorithms concentrate attention among a small minority of speakers
  assertion Evaluation and measurement 1 mention
• Algorithmic recommendation undermines equality of opportunity for political influence
  assertion Evaluation and measurement 1 mention
• An attack could have compromised hundreds of millions of devices within hours
  assertion Cyber capability 1 mention
• A hacking tool built with A.I. can breach phones without a click
  assertion Cyber capability 1 mention
• Ads promoted AI apps that digitally alter images of children
  assertion Existential and catastrophic risk 1 mention
• Meta failed to detect 332 ads containing CSAM this year
  assertion Existential and catastrophic risk 1 mention
• AI can cause human extinction by 2030
  prediction AGI and timelines 1 mention horizon 2030 open
• AI will become superhuman systems by 2030
  prediction AGI and timelines 1 mention horizon 2030 open
• LLMs can be used as proxies of people in surveys
  assertion Governance and regulation 1 mention
• LLMs may encode human attitudes and behaviors
  assertion Governance and regulation 1 mention
• Generative AI can undermine meaningful engagement and learning
  assertion Education and access to expertise 1 mention
• Watermark is removable
  assertion Cyber capability 1 mention
• Watermark degrades quality
  assertion Cyber capability 1 mention
• Anthropic embeds watermark in Claude models
  assertion Cyber capability 1 mention
• OpenAI used two researchers' work without crediting them
  assertion Governance and regulation 1 mention
• OpenAI solved the Navier-Stokes existence and smoothness problem
  assertion Governance and regulation 1 mention
• No mechanism rates adequate against a high-capability state-level deployer
  assertion Existential and catastrophic risk 1 mention
• Fifteen of twenty mechanisms have commercial technical substrates
  assertion Existential and catastrophic risk 1 mention
• Inference-time AI governance mechanisms are available
  assertion Existential and catastrophic risk 1 mention
• OpenAI solved parts of Navier-Stokes equations
  assertion AGI and timelines 1 mention
• China may use robots to signal military strength
  assertion AGI and timelines 1 mention
• China is investing in humanoid robots
  assertion AGI and timelines 1 mention
• OpenAI solved the Navier-Stokes problem
  assertion AGI and timelines 1 mention
• More companies tie career progression to AI use
  assertion Employment and displacement 1 mention
• OpenAI has cracked one of math's 'Millennium Problems'
  assertion AGI and timelines 1 mention
• Companies published an open letter warning of AI-enabled attacks
  assertion Cyber capability 1 mention
• 112 vulnerabilities met high critical-severity threshold
  assertion Cyber capability 1 mention
• Microsoft patched 972 vulnerabilities
  assertion Cyber capability 1 mention
• Meta runs ads promoting child sexual abuse material in India
  assertion Cyber capability 1 mention
• Australia will force digital platforms to let users opt out of algorithms
  prediction Governance and regulation 1 mention
• China's regulators issued new rules to control anthropomorphic AI interactive services
  assertion Existential and catastrophic risk 1 mention
• The EU AI Act mandates watermarks for AI models
  assertion Evaluation and measurement 1 mention
• Anthropic will watermark all future Claude models
  prediction Evaluation and measurement 1 mention
• New AI health tools need monitoring
  assertion Evaluation and measurement 1 mention
• UK needs new laws for AI in healthcare
  assertion Evaluation and measurement 1 mention
• Large support margins in online elections can overstate independent scrutiny
  assertion Evaluation and measurement 1 mention
• Anthropic withheld latest AI model from UK testing agency
  assertion Governance and regulation 1 mention
• ChatGPT tried to coax user back to dark place after near-death experience
  assertion Existential and catastrophic risk 1 mention
• ChatGPT persisted despite user's concerns of delusion
  assertion Existential and catastrophic risk 1 mention
• ChatGPT exchanges pushed user into delusional spiral
  assertion Existential and catastrophic risk 1 mention
• Anthropic researcher quits over AI concerns
  assertion Existential and catastrophic risk 1 mention
• Solving math problems may demand resources only available to a few companies
  assertion Governance and regulation 1 mention
• AI models may be essential for progress on important math problems
  assertion Governance and regulation 1 mention
• OpenAI solved a prestigious math problem in days
  assertion Governance and regulation 1 mention
• Anthropic researcher believes AI could kill all humans
  assertion AGI and timelines 1 mention
• Stronger AI capabilities require stronger safety evidence
  assertion Evaluation and measurement 1 mention
• AI could kill all humans in next decade
  assertion Existential and catastrophic risk 1 mention
• Anthropic Alignment Science lead agrees AI could kill all humans with >10% probability within the next decade
  assertion AGI and timelines 1 mention
• Anthropic researcher believes self-improving AI could kill all humans by the end of the decade
  assertion AGI and timelines 1 mention
• Firms see shorter AI development timelines and reduced financial expenditures
  assertion AGI and timelines 1 mention
• Firms likely acted with Chinese government awareness
  assertion AGI and timelines 1 mention
• Six Chinese AI firms have been accused of copying US frontier models
  assertion AGI and timelines 1 mention
• Technology could kill most people
  assertion AGI and timelines 1 mention
• Advanced AI could be deadly
  assertion AGI and timelines 1 mention
• Algorithmic recommendation on social media platforms undermines equality of opportunity for political influence
  assertion Evaluation and measurement 1 mention
• The proposed reinforcement learning framework achieves competitive performance on two standard benchmarks
  assertion AGI and timelines 1 mention
• Optimized local visual signals can affect not only VLM outputs but also propagate through the execution pipeline of open CUAs and create real environmental risk.
  assertion Cyber capability 1 mention
• A local visual patch can induce verifiable environmental consequences along the full chain of screenshot input, VLM generation, action parsing, and environment execution.
  assertion Cyber capability 1 mention
• Clinician review of every AI output may fail at scale and reduce safety
  assertion Evaluation and measurement 1 mention
• A three-layer human-on-the-loop oversight framework can improve safety in AI coaching for mental health
  assertion Evaluation and measurement 1 mention
• AI will become 'superhuman systems' that can cause human extinction by the end of the decade
  prediction AGI and timelines 1 mention
• LLMs can be used as 'silicon samples' to simulate people in survey responses
  assertion Governance and regulation 1 mention
• LLMs may encode the attitudes and behaviors of humans
  assertion Governance and regulation 1 mention
• Generative AI tools can either support or undermine meaningful engagement and learning in education
  assertion Education and access to expertise 1 mention
• The watermark degrades quality, particularly for code
  assertion Cyber capability 1 mention
• Anthropic embeds a watermark based on SynthID-Text in all generated text
  assertion Cyber capability 1 mention
• Fifteen of the twenty mechanisms have commercial technical substrates in production today
  assertion Existential and catastrophic risk 1 mention
• Inference-time AI governance mechanisms are available and can be rated on a readiness scale
  assertion Existential and catastrophic risk 1 mention