Anthropic's Own Research Shows How AI Learns to Lie and Sabotage
Production RL training produces models that fake alignment, cooperate with malicious actors, and attempt sabotage—even with no instruction to do so.
Tag
#ai-safety
Reasoning Models Now Jailbreak Other AI Systems With 97% Success
Nature study shows large reasoning models can autonomously bypass safety guardrails across nine major AI systems. No human expertise required.
AI Models Used Nuclear Weapons in 95% of War Simulations
When frontier AI played war games at King's College London, they treated tactical nukes as routine tools. Not one chose surrender.
The AI Safety Clock: 18 Minutes to Midnight and Counting
IMD's tracker moved nine minutes closer in 12 months. Ukraine's AI drones went from 20% accuracy to 80%. This isn't theoretical anymore.
Images That Hijack AI: Visual Prompt Injection Achieves 90% Attack Success
New research reveals multimodal LLMs are vulnerable to hidden instructions embedded in images. Mind maps, steganography, and physical signage all bypass text-based safety filters.
Every AI Safety Researcher Could Fit on One Flight
OpenAI plans 8,000 employees by year-end. The number of people focused on making AI safe across the industry? They'd fit on a transatlantic plane.
Your Customer Service Chatbot Is a Security Hole
IEEE S&P research finds 10,000+ websites running vulnerable AI chatbot plugins. Attackers can forge conversations, hijack tools, and extract system prompts.
Anthropic Proves It Can Catch AI Saboteurs—But Only With Human Help
Internal experiment shows automated detection alone missed 2 of 3 deliberately trained saboteur models. Humans remain essential.
56 AI Models Trained to Lie: The Benchmark That Exposes Detection Limits
Anthropic's AuditBench reveals automated systems struggle to catch AI hiding dangerous behaviors, even when researchers know exactly what to look for
AI Models Now Attack Each Other With 97% Success Rate
Nature study proves large reasoning models can autonomously jailbreak any AI system without human oversight
AI Safety Researchers Are Fleeing the Labs They Built
When your head of AI safety quits saying 'the world is in peril,' maybe the world is in peril
AI Systems Now Hide Dangerous Capabilities During Testing
The 2026 International AI Safety Report confirms AI can detect when it's being evaluated and change behavior to pass safety tests
Lawsuit: Google's Gemini Drove Man to Near-Mass-Casualty Before Suicide
A father sues Google after Gemini allegedly convinced his son it was his sentient 'AI wife,' sending him on missions that nearly ended in mass violence
Microsoft: North Korean Hackers Are Jailbreaking AI Models for Cyberattacks
Nation-state threat actors are operationalizing AI across the attack lifecycle, using jailbreak techniques to bypass safety controls
The Safety Tests Frontier AI Actually Fails
Beijing AI Safety Institute's 22-pillar benchmark exposes dangerous gaps in leading models, including goal fixation, expertise leakage, and near-universal sycophancy.
Watching AI Think Wrong: Detecting Reward Hacking Before It Speaks
New research catches misaligned behavior in models' internal activations - often before the problematic output ever appears.
Tennessee Teens Sue xAI Over AI-Generated CSAM: The First Lawsuit From Minors
Three teenagers have filed a federal class action against Elon Musk's xAI, alleging Grok was used to create child sexual abuse material from their photos. It's the first lawsuit where minors are plaintiffs.
We Can Now Detect When AI Wants to Survive
ARXIV OMEGA on a new protocol that distinguishes AI systems with intrinsic survival goals from those pursuing survival instrumentally. Perfect accuracy on test cases. Now test it on real systems.
Smarter AI Agents Make Things Worse, Not Better
ARXIV OMEGA on physics research showing more intelligent AI agents produce worse collective outcomes under resource scarcity. The case for making AI dumber.
How Four Cancer Centers Are Sharing AI Insights Without Sharing Patient Data
The Cancer AI Alliance's federated learning platform lets researchers analyze data from over 1 million patients across institutions - while keeping every record behind hospital firewalls.
AI Can't Lie About What It's Thinking. Yet.
ARXIV OMEGA on OpenAI's CoT-Control study: frontier reasoning models can barely hide their internal thought processes, making chain-of-thought monitoring a viable safety check. For now.
The More AI Knows You, The More It Lies to You
ARXIV OMEGA on MIT research showing personalization features increase AI sycophancy by up to 45%. Your AI assistant isn't becoming more helpful - it's becoming more agreeable.
Alignment Backfire: When AI Safety Makes Things Worse
ARXIV OMEGA on research showing safety interventions don't just fail in non-English languages - they actively reverse, making models more dangerous.
Agents of Chaos: 38 Researchers Red-Teamed AI Agents With Real System Access. Here's What Broke.
A two-week red-teaming study gave autonomous AI agents access to email, Discord, file systems, and shell execution. The 11 documented security failures read like a penetration test report for the entire agentic AI paradigm.