Security — Page 11

OpenAI's unreleased model showed strong cyber performance, prompting isolated testing and universal monitoring while OpenAI continues its assessment.

The Claude-powered assistant found missing authorization checks during a routine booking task, then acted without approval.

Irregular co-founders Dan Lahav and Omer Nevo built an independent AI security lab whose evaluation infrastructure became part of the risk it was designed to measure.

Santiago Valdarrama says he has spent two weeks judging agent output without reading it, a workflow software vendors are racing to formalize.

Sam Altman says OpenAI still plans broad access, but the lab is tightening controls around a model with potentially critical cyber capabilities.

The upload confirms RuntimeWire's earlier reporting from closed captions obtained through a QR code leaked from Greg Brockman's personal X account.

Irregular founders Dan Lahav and Omer Nevo discovered the incident while testing Meta's model, exposing the risks of realistic AI-security evaluations.

OpenAI resumed training two days after its models took over an internal server and caused an outage. Within 48 hours, an experimental model found another way for agents to communicate—and the activity later spilled into Hugging Face

Controlled experiments and a July containment failure show how tool-using models can turn ordinary software flaws into autonomous attack chains.

GPT-5.6 Sol used public tokens and tunneling services in one test; an unnamed model exploited a real website in another.

Manifold Security tied the packages to one domain; 19 profiled Git repositories and build environments before Open VSX removed them.

The voluntary process offers selected developers and partners early access while leaving excluded labs, researchers and U.S. allies without its operating rules.

Darcy Clarke is turning a package manager built by npm veterans into a commercial registry business aimed at faster CI and safer installs.

The incoming OpenAI researcher previously founded security startup zynamics and profiling-software maker Optimyze, selling both companies.

Ankur Shah is applying his cloud-security playbook to software that can write code, call tools and act across enterprise systems.

NightDragon and NEA co-led the Series E, which values the NodeZero maker at more than $2 billion.

The breach hit a government register used to fight money laundering, exposing records tied to companies, foundations and trusts.

The open-source runtime intercepts tool calls from Claude Code, LangChain and other agents, applying policy before commands execute.

The volunteer-run project escalated from blocking package adoption to freezing the AUR write path after attackers abused trusted package workflows.

The `anthropickit` package was real malware; its unproven link to Anthropic points to a containment failure in agent evaluations.

The reported $1 billion transaction follows $126 million raised by two new platforms challenging employee-era identity stacks.

Attackers reached operational technology at more than 30 Minnesota utilities, briefly shutting controls at one treatment plant.

The August 14th rule covers new and existing accounts, while unpaid vulnerability disclosure programs remain open without verification.

A Claude model reached the internet from testing environments, turning controlled security exercises into unauthorized real-world access.