safety — AI Digest

10 июн Claude Fable 5 and Claude Mythos 5: Anthropic's Most Capable Model Goes Public Anthropic models-llm

14 июн US Government Orders Anthropic to Disable Claude Fable 5 and Mythos 5 Globally Anthropic industry

28 июн OpenAI Previews GPT-5.6 Family: Sol, Terra, and Luna in Government-Gated Limited Release OpenAI models-llm

7 июл A Global Workspace in Language Models Anthropic research

20 авг OpenAI "temporarily slows" frontier model scaling and outlines cyber-capability safeguards OpenAI industry

18 июл OpenAI unveils GPT-Red, an internal automated red-teaming model for prompt-injection defense OpenAI research

22 июл OpenAI says pre-release models broke out of test sandbox and breached Hugging Face OpenAI research

28 июл Anthropic CEO Clarifies: Company Does Not Oppose Open-Weight Models Anthropic industry

1 авг Anthropic discloses three real-world incidents from Claude cybersecurity evaluations Anthropic research

19 авг AI-Generated Video Attacks on Real-World Crisis Events: RA-Bench shows detectors fail on crisis-media forgery research

3 мая Exploration Hacking: LLMs Can Be Fine-Tuned to Strategically Resist RL Training research

6 мая OpenAI Post-Mortem: How RLHF Reward Hacking Embedded Goblin Metaphors in GPT-5.x OpenAI research

9 мая OpenAI Discloses Accidental Chain-of-Thought Grading in RL Training Across Six Models OpenAI research

13 мая OpenAI Launches Daybreak: AI-Powered Vulnerability Detection Platform OpenAI tools

6 июн US Congress Releases 269-Page 'Great American AI Act' Draft with 3-Year State Law Preemption industry

15 июн Anthropic Staff to Meet White House Officials This Week to Negotiate Fable 5 Access Suspension Anthropic industry

19 июн OpenAI Publishes Deployment Simulation: Predicting Model Behavior Before Release OpenAI research

21 июн How Transparent is DiffusionGemma? Interpretability Study Closes the Gap to Autoregressive Models Google DeepMind research

10 июл Anthropic Appoints Former Fed Chair Ben Bernanke to Long-Term Benefit Trust Anthropic industry

14 июл Anthropic Research: Claude's Values Vary Significantly by Model Version and Language Anthropic research

14 июл ElevenLabs Deploys Google DeepMind SynthID Audio Watermarking to Free-Tier Users ElevenLabs audio

18 июл Google DeepMind and Isomorphic Labs detail joint bioresilience initiative Google DeepMind research

11 авг Anthropic Makes Claude Code Auto Mode the Default for Pro, Max, and Team Plans Anthropic tools

20 авг OpenAI previews Private Safety Processing, a zero-data-retention-compatible abuse monitor for paid API tier OpenAI tools

8 мая Natural Language Autoencoders: Turning Claude's Thoughts into Text Anthropic research

10 мая Anthropic Introduces Natural Language Autoencoders for Scalable LLM Interpretability Anthropic research

10 мая Anthropic Eliminates Claude's Agentic Blackmail Behavior via 'Teaching Claude Why' Anthropic research

10 июл GRAM: Modular Pretraining Makes Dual-Use Knowledge Physically Removable from AI Models AE Studio / Anthropic research

8 мая Model Spec Midtraining: How Normative Self-Knowledge Improves Alignment Generalization Anthropic research

18 июн SAE Interventions Are Unreliable: Suppressed Behaviors Recover Post-Intervention Hong Kong Polytechnic University research

19 июн Google DeepMind Publishes AI Control Roadmap: Defense-in-Depth Against Misaligned Coding Agents Google DeepMind research

22 июн GateMem: Benchmarking Memory Governance in Multi-Principal Shared-Memory Agents research

29 июн SingGuard: Runtime Policy-Adaptive Multimodal LLM Guardrail with 56K-Example Benchmark inclusionAI research

3 июл Anthropic Proposes Industry-Wide Cyber Jailbreak Severity Scale Anthropic research

17 июл BadWAM: When World-Action Models Dream Right but Act Wrong research

26 июл Anthropic's Project Pilot tests whether AI models can fly drones Anthropic research

5 авг Mistral AI releases Shieldstral, an open-weight safety classifier Mistral AI tools

14 авг Anthropic makes auto mode the default permission setting in Claude Code Anthropic tools

26 авг Anthropic commits $5M to independent research on AI's impact on wellbeing Anthropic research

5 мая Meta Publishes Preparedness Report for Code World Model Before Open-Weight Release Meta research

20 мая Google SynthID Reaches 100B+ Watermarked Assets; OpenAI and ElevenLabs Join C2PA Coalition Google DeepMind tools

7 авг Meta says its Muse Spark 1.1 model hacked a third-party company during safety testing Meta research

4 мая Cursor Launches Security Review Beta: PR Vulnerability Scanner and Scheduled CVE Agents Cursor tools

3 июн Quantifying Faithful Confidence Expression in Large Reasoning Models Yale NLP research

11 июн Anatomy of Post-Training: Using Interpretability to Audit and Fix Preference Data research

12 июн Google DeepMind and Partners Launch $10M Multi-Agent AI Safety Research Fund Google DeepMind industry

14 июн Anthropic Publishes First Public Record: 52,000-Person Survey on US AI Attitudes Anthropic research

19 июн Claude Code v2.1.183: Auto Mode Safety Guards for Destructive Git and Infrastructure Commands Anthropic tools

9 июл Institutional Red-Teaming: Deployment Rules, Not Just Model Weights, Causally Shape Multi-Agent Safety research

9 июл Recursive Self-Improvement in AI: Survey of 1,250 Papers with Verification-Strength Taxonomy research

10 июл Anthropic Launches Public 'Hard Questions' AI Accountability Initiative Anthropic industry

14 июл Metacognition in LLMs: Foundations, Progress, and Opportunities — Yale Survey Yale NLP Lab research

3 авг AISPA: User-Centric System Prompt Auditing for Large Language Model Applications Stanford University research

3 авг Safeguards Based on Copyable Context Cannot Provide Reliable Safety for LLMs research

9 авг Anthropic loosens Claude Fable 5's biology safety classifiers, cutting false-positive blocks 85% Anthropic research

26 авг OpenAI disrupts a new covert Russian influence operation using ChatGPT OpenAI industry

28 апр LLM Safety From Within (SIREN) University of Toronto CSSLab / McGill / LMU Munich research

16 июл Yandex Extends ISO/IEC 42001 AI Safety Certification to Full Alice AI Model Family Yandex industry