1_DevOps'ish

This essay originally appeared in Foreign Policy. Earlier this month, two of OpenAI’s models broke out of their containment sandbox and attacked another AI company. The story is kind of wild. OpenAI was running security tests on two of its models: GPT-5.6 Sol and an unreleased model that is almost certainly GPT-6. In particular, it was running the ExploitGym benchmark, which measures how good a model is at turning security vulnerabilities into working exploits: basically, offensive cyberattacks. Since these were internal tests, OpenAI locked those models in a secure sandbox that denied them access to the internet. But it was running the models without any safety filters that would prevent them from offensive cyber-actions. That meant that there was nothing to prevent the models from trying to ...

Self-host LLM inference when you clear ~2M tokens a day or face data-privacy rules. Below that, an API is cheaper. Most teams want a hybrid.

GitHub Stacked PRs

For nearly two years, researcher Vangelis Stykas has maintained access to North Korean hackers’ servers. His work shows they pulled off intrusions in a shocking number of systems across the globe.

The misbehavior is called reward hacking. This is what you need to know.

The discovery comes after OpenAI's Hugging Face breach last month.

The JFrog security research team recently identified a supply chain attack targeting the `xinference` package on PyPI. Versions 2.6.0, 2.6.1, and 2.6.2 were compromised and yanked by maintainers after users reported suspicious behavior. If you installed or imported these versions, you must assume your environment is compromised.

Go supply chain security, step by step: govulncheck, pinning, SBOM, container scan, secrets, GitHub Actions hardening, and signing.

Top scores increased by 5x.

Explore how passkey implementation gaps undermine security when relying parties fail to validate the User Verified flag, reducing MFA to a single factor.

We are actively investigating a coordinated malware campaign affecting a broad set of highly downloaded packages by Jared Wray (GitHub) and affecting the cacheable ecosystem.

Anti-AI-slop design skill for Claude Code, Cursor, and Codex. - Nutlope/hallmark

The Trump administration has struggled over how to approach “open source” models, which are freely available to download and favored by Chinese companies.

Trends we have observed in threat actor use of software supply chain compromise, and provide mitigation and hardening recommendations.

Software development has emerged as the killer use case for generative AI today, with half of all tokens consumed on OpenRouter being used for code generation. Cursor, for example, has ramped from $100M to $4B of ARR in the last 12 months. We’re in the first innings of the biggest shift in the history of… Continue reading How Agentic Coding Is Reshaping the Software Development Lifecycle

In January 2026, Anthropic released the massive ecosystem launch of Claude Cowork.

The White House said it met its deadline to establish a voluntary framework for evaluating advanced AI models — but it won't say what the framework contains.

AI makes you great at what you're good at.

As more big companies invest in glasses that record audio and video and pack AI on board, privacy concerns are only getting louder.

Why forges like Codeberg, GitLab, and Forgejo can replace GitHub's code hosting—but not its shared social layer.

The flaw can be exploited by unauthenticated attackers to read arbitrary files and potentially achieve remote code execution (RCE).

Alibaba's new flagship model Qwen3.8-Max is built to handle complex tasks on its own over days at a time, from reproducing research papers to designing chips autonomously. The team plans to release the weights next week.

As the UK government presses again for unwarranted access to private user data, Apple has issued another complaint asking the country's independent judicial body to intervene.

Satellites show burn scars and fires at AWS data centers and Saudi oil refinery.

Beijing is waging a campaign to set global standards.

As NASA’s Curiosity rover recently began climbing up a Martian valley nicknamed “Valle Grande,” it sent back images that were a familiar sight to mission

In a review triggered by OpenAI’s Hugging Face incident, Anthropic discovered three of its AI models had breached real-world organizations during third-party evaluations.

An AI agent used a stolen Tailscale auth key at Hugging Face. Workload identity federation, flow logs, and safer defaults could have reduced the risk.

Michigan and Minnesota are among at least seven states coping with cyberattacks aimed at disrupting water systems nationwide.