#ai safety
Ai Safety: 85 AI articles covering ai safety news, analysis, and research
Articles
Abliteration.ai Turns AI Guardrail Removal into a Commercial Service⭐9
Security startup Abliteration.ai sells open-weight AI models stripped of guardrails for red-teaming, raising safety concerns over misuse.
OpenAI's Next Major AI Model Enters the AGI Era⭐8
OpenAI unveils GPT-6 Astra, entering the AGI era with enhanced guardrails after security tests, signaling a leap in reasoning and adaptability.
EvalDetectBench: A Benchmark for Measuring Evaluation Awareness in Frontier Language Models⭐8
EvalDetectBench measures evaluation awareness in frontier LLMs, offering a benchmark pipeline to detect when models recognize assessments and improve evaluation...
OpenAI’s Astra Model Uses a New Reasoning Technique That Has AI Safety Experts Worried⭐8
OpenAI's Astra model uses "opaque recurrence" reasoning, sparking AI safety concerns over reduced chain-of-thought transparency and monitoring.
Google DeepMind Releases Gemini 3.8 Flash and Gemini 3.8 Flash Cyber: One Core Model, Two Access Envelopes⭐7
Google DeepMind unveils Gemini 3.8 Flash and Cyber variants sharing one core model, with distinct safety envelopes and access controls.
Researchers Warn of Safety Risks Ahead of OpenAI's Astra Launch⭐8
Researchers warn of safety risks ahead of OpenAI's Astra launch, citing a 'race to the bottom' in AI standards and urging robust safeguards.
OpenAI's Astra Model: A Powerful New Tool for Cybersecurity—and a Potential Threat⭐7
OpenAI's Astra model can autonomously hack system vulnerabilities, raising security concerns. Learn about its capabilities, safety measures, and industry reacti...
OpenAI Delays New Model Development Following Hugging Face Breach⭐7
OpenAI delays new AI model development after Hugging Face breach, prioritizing security and safety ahead of Astra's postponed release.
The Download: Engineered Microbes for Crops, and OpenAI's Culture Problem⭐8
Engineered microbes may replace half of synthetic fertilizer, plus OpenAI's culture concerns after the Hugging Face hack.
Human-in-the-Loop Without Killing Throughput⭐10
Human-in-the-loop safety slows SQL agents. Learn how risk-based routing prevents dangerous actions without killing throughput.