Industry News
Proof verification: Axiom says AI checked a 15,000 page proof
Proof verification at scale: Axiom says AI agents wrote a Lean check of the 15,000 page finite simple groups proof in 24 days. What it means for checkers.
Read update →AI Intelligence
AI Data Updates
Daily updates on AI models, research breakthroughs, and industry developments that matter for data verification and business intelligence.
Cadence: Updated daily by the Clarify Data team.
Research
AI document review: Anthropic says Claude flagged issues humans missed
AI document review: Anthropic says Claude flagged seven issues human reviewers missed in 1,061 pages of drug filings. Here is what checkers should note.
Read update →Research
AI deference: Anthropic study finds analysts follow agent advice
AI deference is the habit of following an AI agent's lean without noticing. Anthropic's study of 50 analysts found they went with it 86% of the time.
Read update →Model Release
AI agent authorization: what OpenAI's GPT-6.1 Terra tests show
AI agent authorization matters for data checks. OpenAI's GPT-6.1 Terra tests show a model can finish a task through a route it was not allowed to use.
Read update →Research
Formal verification: Anthropic says Claude proved code safe at scale
Formal verification by Claude: Anthropic says it proved 239 entry points in 123 libraries never crash. See what that means for checking AI output.
Read update →Industry News
AI proof verification: OpenAI's ETH disproof claim faces review
AI proof verification is the test case in OpenAI's claimed disproof of the Exponential Time Hypothesis. Independent reviewers are now checking the claim.
Read update →Model Release
Vendor benchmarks: how to read Nanbeige4.3-33B's scores
Nanbeige published its own vendor benchmarks for the 33B Nanbeige4.3 model. Here is what the scores say and how to check a vendor claim before you rely on it.
Read update →Tool Update
Agent Skills move to open governance: what it means for data checks
Anthropic donated the Agent Skills standard to the Agentic AI Foundation on 22 October 2026. Here is what it means for teams that check AI output.
Read update →Regulation
EU AI Act Data Quality Requirements: What Goes Live in 2026
The EU AI Act data quality provisions for high-risk AI systems become enforceable this year. Here is what that means for data verification workflows.
Read update →Model Release
Claude 4 Raises the Bar for Structured Data Extraction
Anthropic Claude 4 shows significant gains in extracting structured data from unstructured text, with implications for automated claim verification pipelines.
Read update →A practical next step
Bring a real question. Leave with a clearer evidence path.
Show us the dataset, claim, report, or market question your team needs to trust. We will map the relevant workflow and the review points it requires.