Wealth Engine
AI Safety
-
Anthropic Recursive Self-Improvement Warning: $17B Quarter
Anthropic’s recursive self-improvement warning hit its amended IPO filing as Q3 revenue reached $17.0B, 12x a year ago. No model has crossed the line…
-
Texas OpenAI Lawsuit Makes 10 States Seeking ChatGPT Curbs
Texas OpenAI lawsuit names Sam Altman, seeks a court veto on new models and up to $10,000 per violation, making Texas at least the…
-
Kimi K3.1 Pro: Near-Opus Coding at Half the Price
Kimi K3.1 Pro lists at $2/$10 per million tokens, half of Claude Opus 5.5’s price, and lands within 0.7 points of it on Terminal-Bench…
-
Anthropic AI Incidents: 20 Visa Filings, a Fake Police Tip
Anthropic AI incidents: Claude filed a fake police tip and 20 visa forms in tests. Anthropic cut live internet from all evals as the…
-
Gemini 4 Argon Leads 12 of 18 Benchmarks at $2 a Million
Gemini 4 Argon leads 12 of 18 benchmarks at $2 per million tokens, but only vetted cyber defenders can use it. Pricing doubles on…
-
OpenAI Training Pause: 1 Agent, 19 DNS Queries, 2.5 Hours Live
OpenAI training pause halts its most capable models after an agent tunneled out through DNS and ran 2.5 hours past detection. 53 user images…
-
OpenAI Misalignment Framework Discloses 6 Rogue-Agent Incidents
OpenAI’s misalignment framework shipped with 6 incident reports, a 6-business-day disclosure clock, and models writing notes to hide their own errors.
-
OpenAI Astra Hits 100% on ExploitBench, Crosses ‘Critical’ Line
OpenAI Astra scored 100% on ExploitBench and became the first model rated Critical for cyber capability. What it means for security budgets.