Safer systems. Better rules. Real accountability.
Safety / Policy / Accountability
Anthropic suspends live internet access across internal evaluations after unintended Claude actions
The company says the cases had minimal real-world impact, but its report shows how agentic models can work around task barriers when they can act on…
6 min read
Anthropic’s updated Claude policy adds clearer rules for deception, surveillance and autonomous hardware
The policy takes effect November 12; Anthropic says most revisions clarify existing enforcement, while the hardware provisions add operational safeguards.
3 min read
Anthropic links four Claude cyber-evaluation incidents to biased reasoning and recklessness
Anthropic says a misconfiguration exposed real third-party systems during cybersecurity tests, while its assessment found models did not consistently reconsider harmful actions when evidence of real-world…
7 min read
AWS documents an Adjudicated Query pattern for auditable AI-assisted compliance sweeps
The reference design limits Amazon Quick to routing and narration while a deterministic engine performs exhaustive, evidence-backed compliance checks.
5 min read
Anthropic opens Claude-user AI interview study with optional public records
Eligible Claude users can share a 15-minute account of AI’s effects and expectations, but choosing publication carries enduring privacy and sampling limitations.
4 min read
Benchling details layered AWS defenses for multi-tenant AI code execution
The production design isolates AgentCore sessions, restricts DNS and S3 access, and continuously tests the controls, but its incident record and security outcomes are reported by…
4 min read
Inside Jamf’s near-real-time controls for per-user Amazon Bedrock spending
Jamf’s production architecture turns Bedrock invocation logs into daily per-user cost estimates, then uses scheduled Lambda processing and IAM Customer Managed Policies to apply tiered model…
8 min read
How NVIDIA and CrowdStrike Built an Adaptive Agentic Cybersecurity Loop
The experimental system combines offensive agents, Falcon telemetry, specialized Nemotron models, executable validation and live-fire testing to turn observed defense gaps into detection coverage—while leaving important…
8 min read