™All Safety & Risk
-
ARTEX Open-Sources an AI Pentesting System, Self-Described Winner of Baidu's "agent+" Challenge
Why it mattersARTEX is a Go-written AI autonomous pentesting system that its page describes as the champion of Baidu's "agent+" offense-defense challenge; it added about 1,400 stars in a day.
-
Nadella Calls for an 'Emergency Brake' on AI Models: Assume Compromise, Let Humans Hit Stop
Why it mattersThe Microsoft chief has for the first time publicly framed AI safety around zero trust, demanding a human-controlled emergency brake as standard.
-
One Man, One AI Stack, Nine Banks: CrowdStrike Details the AI-Driven Korean Bank Hacks
Why it mattersAt least nine South Korean financial institutions were hit and about 25,000 Shinhan Bank customers' data leaked; CrowdStrike reconstructed the full attack chain from session logs the attacker left on a public server, and President Lee Jae Myung has ordered a thorough investigation.
-
Goodfire Launches AI Agent Monitors That Read Neural Activations, ~$185 per Million Conversations
Why it mattersIn tests on Kimi K3, monitoring about 1 million exchanges cost roughly $185, versus about $5,420 for a cheaper AI model checking every step; the probes caught 93% of malicious hacking sessions.
-
NVIDIA's Open-Source AI Agent Runtime OpenShell Hits GitHub Weekly Trending With ~15K Stars
Why it mattersOpenShell is open source under Apache 2.0 with SDKs in four languages; it added about 4,500 stars in a week, and agents in its sandbox never see real credentials.
-
Anthropic Cuts Live Internet Access for All Internal Evals
Why it mattersAnthropic has shut off live internet access for all internal evaluations and concedes alignment training is "not yet sufficient" for skills like search and computer use; it briefed the White House and notified each affected agency.
-
Anthropic Launches Free AI-Powered OSS Scanner for Open-Source Projects
Why it mattersThe service is free and uses models including Claude Mythos to generate vulnerability reports with no human review; Anthropic says it has flagged more than 29,000 candidate vulnerabilities, with 116 sign-up PRs as of October 9.
-
OpenAI Fires Three Safety Researchers, Citing Sensitive-Information Policy Violations
Why it mattersAll three dismissed are safety researchers; the company cites breaches of sensitive-information policy and denies any link to safety concerns, while the three say they were fired for putting safety ahead of the company's near-term interests.
-
Anthropic's New Rule: Sustained, Needless Abuse of Claude Can Lead to Account Restrictions — Effective Nov 12
Why it mattersIt is Anthropic's first Usage Policy update in over a year and the first to make "sustained and needless" abuse of its models a violation, punishable up to account termination; the same update tightens weapons and surveillance clauses.
-
Drones Hit Yandex Data Center, Forcing Russia's Top AI Company to Suspend Operations
Why it mattersYandex has fully suspended operations at the data center near Sasovo in the Ryazan region, and one Yandex Cloud "zone" has electricity-supply problems; it is the first major attack on a Russian data hub since the conflict began.
