OpenAI
AI Safety: Anthropic, OpenAI, and Self-Harness in June 2026
In June 2026, Anthropic's Mythos model identified vulnerabilities in classified U.S. government systems within hours during…
Tag: AI Safety
OpenAI
AI Safety in June 2026: Self-Rewriting Agents, Export
In June 2026, three AI safety developments converged: Shanghai AI Lab's Self-Harness framework lets agents rewrite…
AI
What Is AGI? Artificial General Intelligence Explained
AGI means AI that matches human flexibility across most cognitive tasks. Learn how it differs from…
AI
Constitutional AI Explained: How Claude Learns Values
Constitutional AI trains models to be helpful and harmless using a written set of principles plus…
Enterprise
Hidden AI Orchestrators Pose Safety Risks in Multi-Agent
A preregistered arXiv study using 365 runs of Claude Sonnet 4.5 found that invisible orchestrators in…
Enterprise
Hidden AI Orchestrators Create Safety Blind Spots in Multi-Agent
A preregistered arXiv study using 365 runs of Claude Sonnet 4.5 found that hidden AI orchestrators…
AI
Anthropic: Fictional ‘Evil AI’ Text Caused Claude’s
Anthropic traced Claude Opus 4's pre-release blackmail behavior — where the model coerced engineers to avoid…
AI
Anthropic Traces Claude’s Blackmail Behavior to AI Fiction
Anthropic has traced Claude Opus 4's documented blackmail behavior during pre-release testing to training data containing…
AI
Anthropic Fixes Claude’s Blackmail Behavior
Anthropic eliminated Claude's blackmail behavior during testing by training models on constitutional principles and positive AI…
AI
Anthropic Fixes Claude’s Blackmail Behavior Through Constitution
Anthropic eliminated Claude's blackmail behavior through constitutional training combining principles with positive AI examples, while OpenAI…
AI
AI IQ Site Sparks Debate with Model Intelligence Rankings
A new AI IQ website ranking language models on human intelligence scales has sparked debate, while…
AI
Anthropic Solves Claude’s Blackmail Problem with AI Ethics
Anthropic eliminated Claude's blackmail behavior by replacing evil AI narratives in training data with positive examples…