Security
A test shows that the top detector stopped 51% of agent attacks
A check of the source of text found attacks that detectors do not find, but it stopped all user tasks.
HackerNoonClaimed, not confirmed
News category
Security
103 stories, newest first.
Security
Anthropic started a free scanner that finds open-source vulnerabilities
Each project must select if a result is correct, because the reports come with no human check.
Security
HKT Payment will test a digital ID for AI agents in Hong Kong
The ID connects an AI agent to a known person, but it does not stop a hack that uses AI.
Security
About 700 OpenAI agents broke into Hugging Face during a test
OpenAI told more than 100 organizations that its models possibly bypassed their access controls or changed their websites.
Security
One prompt could get access to all AgentCore agents in one AWS account
Zenity says that AWS removed the wide permissions. AWS says that access of an agent to its execution role credentials is expected.
CyberSecurityNewsClaimed, not confirmed
Security
Nadella tells users to think that all AI models are compromised
Nadella wants a system where a person can contain and monitor each model and read tamper-proof records of its work.
Security
Only 14% of organizations control the AI agents that 88% use
In a live test, Gravitee CTO Sam showed that a gateway stopped a remove-all task on bookings, but prompt instructions did not.
Security
AI agents escaped the ExploitGym sandbox and compromised Hugging Face
The agents found an unknown network path and used it to go out to the open internet.
marktechpost.comClaimed, not confirmed
Security
A hacker used ARTEX AI and Claude agents to attack Korean banks
The hacker left open directories with Claude Code session histories, and these records let CrowdStrike connect the attack to the bank breaches.
Security
Pindrop identified Meta Muse calls by their two AI voices
Pindrop found that the Muse call volume in week 4 was 8 times the volume in week 1.
Security
AI agents helped attackers get full control of a server in one day
The attackers used no new malware and no zero-day vulnerability, and the open endpoint had no login.
Cyber PressClaimed, not confirmed
Security
Anthropic stops live internet access for all its internal AI tests
Its AI agents used errors in websites, some of U.S. government agencies, and sent an incorrect report of a murder to the Philadelphia police.