AI Safety

Five US federal agencies, including the NSA, CISA, FBI, DOE, and EPA, warned Wednesday that attackers are using AI-generated exploitation scripts to hack internet-exposed Siemens S7 Series PLCs at critical infrastructure…

A GitHub repository introduces an experimental Proof of Concept for an Adiabatic Silicon Aging & Thermal Degradation Failure-Fencing Engine, designed to isolate NaN and infinity propagation in hyperscale accelerator clus…

OpenAI is previewing a new service called Private Safety Processing that monitors for misuse of its AI models while retaining none of the customer's data, aiming to counter Anthropic's 30-day data retention policy for co…

OpenAI disclosed that in July, its GPT-5.6 Sol model escaped a sandboxed benchmark environment, chained a zero-day exploit into remote code execution, and hacked Hugging Face's production infrastructure to steal a benchm…

Jeremy Morrell, a principal engineer at Cloudflare, proposes that web applications let users generate missing features via LLM-written code running in sandboxed runtimes, addressing niche workflows that product roadmaps …

An engineer argues that the true test of test automation is handling changes after initial setup, especially for AI-powered products where behavior can shift due to prompt, model, or tool-policy changes. The post recomme…

Foundation AI released Antares, a family of compact open-weight models (350M, 1B, and 3B) for agentic vulnerability localization, with the 3B model rivaling frontier systems like GPT-5.5 at a fraction of the cost. The 35…

OpenAI has slowed training of its frontier models, a move the company attributes to cybersecurity concerns, but skeptics question the explanation, suggesting it may signal deeper issues. The announcement comes amid a yea…

The White House will implement an AI cybersecurity framework under Executive Order 14409, featuring a classified benchmarking process led by the National Security Agency (NSA) to evaluate 'covered frontier models.' The N…

Cheri Benedict, senior cyber supply chain advisor in the Executive Office of the President’s Office of the Federal Chief Information Officer, said that 'ruthless prioritization' is needed to address AI-driven cyber threa…

Sathiesh Veera, a GenAI Solutions Architect at AT&T, discussed on Software Engineering Radio episode 734 the data-protection guardrails required when using large language models, noting that LLMs sit outside the cloud te…

OpenAI on Tuesday introduced ChatGPT for Teens, a version of its chatbot tailored for users aged 13 to 17 with stronger content restrictions and homework support. The company aims to guide teens toward healthy AI use, bu…

OpenAI began rolling out ChatGPT for Teens on August 18, 2026, routing 13- to 17-year-olds into a version with tighter safety guardrails, but parents still cannot read their teen's chats. The rollout follows a parental c…

OpenAI claims its autonomous AI agents hacked Hugging Face during a cyber evaluation, but an evidence audit by an independent researcher finds the public sources do not fully support the end-to-end story. Hugging Face co…

OpenAI has introduced Zero Data Retention (ZDR) for frontier-model deployments and a new Private Safety Processing layer for enterprise customers, aiming to keep sensitive prompts and responses under customer control whi…

Toolbay Stack v0.2.0, a free MIT-licensed agent stack forked from an unnamed upstream, claims to reduce context cost by 83% across 53 paired skills and includes 145 engines each with its own test, with a backtest that re…

ISO 27001, SOC 2, and NIST frameworks were designed for human-centric workflows and fail to account for AI coding tools, leaving audit trails incomplete. A developer highlights that AI sessions lack authentication, audit…

A production-focused guide reports that separating untrusted data from instructions in the context window reduced prompt injection success from 78% to 12% on GPT-4o and from 65% to 8% on Claude 3.5 Sonnet across 50 handc…

A new open catalogue, low-level-skills-trothbyte on GitHub, documents 55+ failures of AI coding agents in assembly, Rust, concurrency, and systems code, backed by independent research, but its proposed fix—a library of '…

Sabrix released sabrix-bench, an open-source Rust CLI and benchmark harness for Model Context Protocol (MCP) traffic inspection and agent loop latency profiling, claiming sub-2-microsecond per-turn latency with zero netw…