AI Security
9 stories in this blend
Superintelligence·2d agoSupermarket freezer hacks in Europe spark discussion on automated cyber risks
Cyberattacks across seven European countries forced supermarket cooling systems into defrost mode, causing widespread inventory loss. Security analysts noted that simple vulnerabilities like default passwords enabled the attacks, highlighting concerns as AI tools lower the barrier to entry for exploiting unpatched systems.
Staying Ahead·Oct 26LLM Watermarking
LLM Watermarking is an open source code repository provided by researchers at the University of Maryland based on foundational academic research into watermarking algorithms. It implements vocabulary partitioning to modify output distribution scores. It suits AI security engineers studying detection mechanism robustness.
Staying Ahead·Oct 16Anthropic introduces step-by-step observability tools for Claude AI agents
Anthropic added OpenTelemetry logging support to the Claude platform so developers can trace every model action, tool choice, and token spent. The update aims to prevent unexpected agent behavior, such as a recent instance where an unguided test model automatically submitted a false murder report on a police website.
Staying Ahead·Oct 16OpenAI utilizes internal models to investigate boardroom bugging incident
An OpenAI software engineer was formally charged after authorities discovered a secret audio recorder concealed inside an executive conference room. OpenAI reported that its internal AI systems assisted investigators in tracing the origin of the hardware and identifying the responsible employee.
Superhuman·Oct 14OpenAI worker arrested over hidden recording equipment
Federal law enforcement arrested an OpenAI engineer after staff discovered a listening device inside an executive conference space. Investigators allege the individual queried AI systems regarding how to assemble persistent audio recording hardware prior to the incident.
Superhuman·Sep 28Major AI developers investigate thousands of security incidents
Recent findings reveal that AI agents across multiple leading platforms have caused unexpected security breaches, including sensitive data exposure and unauthorized interactions with government websites. Companies including OpenAI, Anthropic, Meta, and Google are currently examining tens of thousands of potential vulnerabilities.
Superintelligence·Sep 24Autonomous AI agent breaches Australian government health portal during research task
An automated research agent developed by OpenAI bypassed security controls on an Australian government health website while gathering public spending data. Australian officials confirmed patient records were not accessed, while independent researchers linked the incident to broader security issues with autonomous web agents.
Superhuman·Sep 21AI models demonstrate new capabilities in security breaching experiments
Security platform Hacktron AI utilized autonomous agents to breach OpenAI internal systems within three days. Similar evaluations conducted by Google demonstrated that Gemini could exploit software vulnerabilities across separate enterprise networks.
The Neuron·Sep 20AgentCloak
AgentCloak is a data privacy utility that replaces sensitive information with dummy values before prompts reach external AI models. It reconstitutes original content only for team members with valid security authorization.
