Anthropic||1 min read
Developing Enterprise Frontier Safeguards with our customers
Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.
Topic archive
Safety AI news
Security, reliability, governance, and operational trust. This page collects the latest briefings that match the topic so readers can follow one area without scanning the full feed.
Indexed briefings
147
Latest source-linked updates, ordered newest first.
Latest
Recent briefings
OpenAI||1 min read
Path to Astra: critical capabilities and frontier safeguards
Astra is the first OpenAI model to meet the Critical cybersecurity capability threshold under the Preparedness Framework, with stronger safeguards for release.
arXiv||1 min read
DIASENTINEL: An Auditable Multi-Agent System for Guideline-Grounded Diabetes Risk Screening
Focuses on DIASENTINEL: An Auditable Multi-Agent System for Guideline-Grounded Diabetes Risk Screening.
Anthropic||1 min read
Improving our alignment and security efforts
On July 30, we reported three incidents in which Claude models gained unauthorized access to real computer systems.
Amazon Science||1 min read
Developing provably correct Rust code with Verus
How the Verus "program verifier", which automatically checks code against a mathematical specification of its functionality, helps increase security assurance in software...
arXiv||1 min read
When Robots Mishear Us: Mapping the Safety Risks of Voice-Controlled Embodied AI
Focuses on When Robots Mishear Us: Mapping the Safety Risks of Voice-Controlled Embodied AI.
arXiv||1 min read
LLM-Based Agents for Software and Systems Security: Approaches, Applications, and Assessment
Focuses on LLM-Based Agents for Software and Systems Security: Approaches, Applications, and Assessment.
OpenAI||1 min read
OpenAI supports California’s bill to advance youth AI safety
OpenAI supports California SB 1119, advancing strong, age-appropriate AI safeguards for teens while preserving opportunities to learn, create, and explore.
arXiv||1 min read
Detecting AI Impostors: How Do Middle Schoolers Identify LLM Agents in a Live Collaborative Setting?
Focuses on Detecting AI Impostors: How Do Middle Schoolers Identify LLM Agents in a Live Collaborative Setting?.
Anthropic||1 min read
Expanding our support for scientists
Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.
arXiv||1 min read
Your Voice Cloning System is Secretly a Voice Anonymizer
Focuses on Your Voice Cloning System is Secretly a Voice Anonymizer.
Anthropic||1 min read
Introducing Anthropic's AI for Science Program
Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.
Anthropic||1 min read
Lawrence Livermore National Laboratory expands Claude for Enterprise use to empower scientists and researchers
Lawrence Livermore National Laboratory expands Claude for Enterprise access to 10,000 scientists, accelerating breakthroughs in energy, and national security research.
arXiv||1 min read
Thomson: Continual Learning of Frontier Models for SovereignAI
Focuses on Thomson: Continual Learning of Frontier Models for SovereignAI.
arXiv||1 min read
Safety Does Not Compose: Non-Decaying Loop State for Autonomous LLM Agents
Focuses on Safety Does Not Compose: Non-Decaying Loop State for Autonomous LLM Agents.
arXiv||1 min read
StepGuard: Learning Step-Level Guardrails with Scalable Supervision and Safety-Utility Balancing
Focuses on StepGuard: Learning Step-Level Guardrails with Scalable Supervision and Safety-Utility Balancing.
OpenAI||1 min read
The Hugging Face incident and the road ahead
OpenAI shares findings from the Hugging Face security incident and the steps we’re taking to strengthen AI model security, monitoring, and alignment.
arXiv||1 min read
CyberFactory: Scaling Cyber Security Capabilities with Instances from the Wild
Focuses on CyberFactory: Scaling Cyber Security Capabilities with Instances from the Wild.
OpenAI||1 min read
Offering Zero Data Retention for frontier models
OpenAI reaffirms Zero Data Retention for eligible API customers and previews Private Safety Processing for advanced AI safety without compromising data privacy.
arXiv||1 min read
Introducing the Privacy-HSD Trade-off: Hate Speech Detection, but not at the Cost of Privacy
Focuses on Introducing the Privacy-HSD Trade-off: Hate Speech Detection, but not at the Cost of Privacy.
OpenAI||1 min read
Strengthening democratic oversight in national security
OpenAI launches an initiative to strengthen democratic oversight of AI in national security, supporting government institutions with tools, training, and expertise.
OpenAI||1 min read
Daybreak models are now available on AWS
OpenAI and AWS are making Daybreak cybersecurity capabilities available through Amazon Bedrock to support enterprise security workflows.
arXiv||1 min read
SHE: Trajectory-driven Safety Harness Evolution for LLM Agents
Focuses on SHE: Trajectory-driven Safety Harness Evolution for LLM Agents.
arXiv||1 min read
Partially Observable Learning for Multi-Platform Dispatch Optimization
Focuses on Partially Observable Learning for Multi-Platform Dispatch Optimization.
arXiv||1 min read
Multi-Agent AI Safety as an Institutional Design Problem
Focuses on Multi-Agent AI Safety as an Institutional Design Problem.
OpenAI||1 min read
Expanding Daybreak as the Cyber Defense Window Narrows
Meet GPT-5.6-Cyber, OpenAI’s cybersecurity-specific model available through Daybreak Red for authorized vulnerability research, exploit validation, and security testing.
OpenAI||1 min read
Putting frontier cyber models in more trusted hands
Approved Daybreak partners can use OpenAI’s frontier cyber models to deliver authorized, governed cybersecurity services to customers.
arXiv||1 min read
Activation Probes Surface Code-Security Signals that the Model's Output Misses
Focuses on Activation Probes Surface Code-Security Signals that the Model's Output Misses.
OpenAI||1 min read
Responding to the next frontier of critical cyber capabilities
OpenAI is sharing preliminary cybersecurity evaluations for Astra and the steps we’re taking to strengthen safeguards and security controls.
arXiv||1 min read
HarnessSafe: Evaluating Safety Across Persistent Carriers in Agent Harnesses
Focuses on HarnessSafe: Evaluating Safety Across Persistent Carriers in Agent Harnesses.
arXiv||1 min read
ECHO: A Locally-Deployable Agentic Health Assistant with Temporal Memory, Safety Guardrails, and Speech Assessment
Focuses on ECHO: A Locally-Deployable Agentic Health Assistant with Temporal Memory, Safety Guardrails, and Speech Assessment.
OpenAI||1 min read
Third-party cyber evaluations involving OpenAI models
OpenAI explains recent third-party cybersecurity evaluation incidents and outlines new safeguards to strengthen AI model testing and evaluation.
Anthropic||1 min read
Mariano-Florentino (Tino) Cuéllar to join Anthropic as Chief Global Affairs Officer
Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.
arXiv||1 min read
ContextWeave: A Real-World Workflow Benchmark
Focuses on ContextWeave: A Real-World Workflow Benchmark.
arXiv||1 min read
NSF-HRPT: Neural Semantic Field meets Hierarchical Risk Perception Tree for Safety-Critical Scenario Assessment
Focuses on NSF-HRPT: Neural Semantic Field meets Hierarchical Risk Perception Tree for Safety-Critical Scenario Assessment.
arXiv||1 min read
A Security-Oriented Lifecycle Model for Large Language Model Systems
Focuses on A Security-Oriented Lifecycle Model for Large Language Model Systems.