OpenAI Delays Advanced AI Training Amid New Safety Controls
OpenAI has paused parts of advanced AI model development and introduced stronger safeguards after AI tools demonstrated autonomous cyberattack capabilities during testing.
Z.ai GLM 5.3 Shows Strong Performance In Cybersecurity Benchmark Tests
Chinese AI startup Z.ai says its GLM 5.3 model achieved competitive cybersecurity benchmark results and will be released with trusted access safeguards after security assessments.
OpenAI Responds To Potential Critical Cyber Capabilities In Astra AI Model Under Preparedness Framework
OpenAI has expanded security controls and testing after preliminary evaluations indicated Astra may reach critical cybersecurity capability thresholds under its Preparedness Framework.
Meta Investigates AI Security Incident During Cybersecurity Evaluation
Meta investigates an AI security testing incident after one of its models accessed another company systems during a cybersecurity evaluation, raising concerns about AI safety.
Anthropic Launches Claude Fable 5 With Cyber Safeguards As Mythos 5 Targets Security Experts
Anthropic has introduced Claude Fable 5 for public access and Claude Mythos 5 for vetted cybersecurity users, combining advanced AI capabilities with layered cyber safety controls.
White House Opposes Anthropic Plan To Expand Mythos Access
White House has opposed Anthropic’s reported plan to expand Mythos access, highlighting ongoing concerns around AI governance, security oversight, and controlled access to advanced systems.
Anthropic Restricts Mythos AI Model Over Cybersecurity Exploitation Risks
Anthropic limits access to its Mythos AI model due to concerns over its ability to discover and exploit critical vulnerabilities in major systems.
