today's signal · no scroll
the brief
Agents stepped into infrastructure: Cloudflare opened its Agents SDK to external frameworks, shipped a Flue integration, and introduced agent-powered Cloudflare One deployments. Open weights advanced with Z.ai’s MIT-licensed 753B GLM-5.2, while Anthropic’s Claude Code delivered practical UX upgrades. New motion forecasting work, agent evals, and a jailbreak of ChatGPT’s image tool rounded out the day.
Mindgard reports a reproducible jailbreak that coaxes ChatGPT’s image tool into violent and sexual outputs, highlighting lingering safety gaps teams should account for in risk reviews.
ChatGPT's image generator can be manipulated to produce violent, sexual content — Article URL: https://mindgard.ai/blog/chatgpt-spontaneously-generated-violent-images-from-a-viral-prompt Comments URL: https://news.ycombinator.com/item?id=48578894 Points: 54 # Comments: 75
Cloudflare makes its Agents SDK a framework-agnostic runtime, debuts Flue as the first integration, and brings agent management into the dashboard, easing deployment on the edge.
Bringing more agent harnesses and frameworks to Cloudflare, starting with Flue — The Agents SDK is now a runtime any agent framework can build on. Today we're opening up the Agents SDK primitives, with Flue as a first framework targeting Agents SDK, and rolling out agents in the dashboard.
New library of agent skills lets AI agents plan, deploy, and manage Zero Trust configurations on Cloudflare without manual migration, turning security automation into repeatable tasks.
Introducing the Cloudflare One stack: agent-powered deployment — The Cloudflare One stack is a library of agent skills that gives any AI agent the knowledge it needs to plan, deploy, and manage a Zero Trust environment — no migration calls required.
Latest release brings in-prompt /config for settings, macOS Apple Events opt‑in in sandbox, a presence file to suppress pushes, Bun 1.4, and better long-paragraph streaming.
v2.1.181 — What's changed Added /config key=value syntax to set any setting from the prompt (e.g. /config thinking=false) — works in interactive, -p, and Remote Control Added sandbox.allowAppleEvents opt-in setting that lets sandboxed commands send Apple Events on macOS Added CLAUDE_CLIENT_PRESENCE_FILE environment variable: point it at a marker file to suppress mobile push notifications while you're at the machine Upgraded the bundled Bun runtime to 1.4 Improved streaming of long paragraphs:...
Z.ai released 753B-parameter GLM‑5.2 text-only weights (1.51TB) under MIT, positioning it as a top open model for non-multimodal tasks and serious offline deployments.
GLM-5.2 is probably the most powerful text-only open weights LLM — <p>Chinese AI lab <a href="https://z.ai/">Z.ai</a> released GLM-5.2 <a href="https://x.com/Zai_org/status/2065704919299235870">to their coding plan subscribers</a> on June 13th, and then yesterday (June 16th) released the full open weights under an MIT license. Similar in size to their previous GLM-5 and GLM-5.1 releases, this is 753B parameter, <a href="https://huggingface.co/zai-org/GLM-5.2">1.51TB</a> monster - with 40 acti...
Midjourney unveiled a dedicated medical offering, signaling deeper domain focus for regulated, expert workflows beyond general creative generation; details are sparse but a signup page is live.
Midjourney Medical — https://www.midjourney.com/medical Comments URL: https://news.ycombinator.com/item?id=48579650 Points: 95 # Comments: 106
Tailscale now exports audit and network logs directly to Azure Blob Storage, simplifying retention and integration with existing SIEM and compliance pipelines on Microsoft stacks.
Send Tailscale logs to Azure Blob Storage — Keep Tailscale logs with the rest of your security data.
AllenAI and Hugging Face present language-guided 3D motion prediction, combining LLM priors with motion models for controllable forecasting—useful for robotics, animation, and embodied agents.
MolmoMotion: Language-guided 3D motion forecasting
New benchmark environment targets agents that iteratively improve themselves, enabling reproducible evaluation of planning, tool-use, and self-modification loops across tasks.
SEAGym: An Evaluation Environment for Self-Evolving LLM Agents — Abstract page for arXiv paper 2606.17546: SEAGym: An Evaluation Environment for Self-Evolving LLM Agents
A multi-source, ATT&CK‑mapped log corpus arrives with small language model evaluations, offering a realistic dataset for training and testing security co-pilots and detection agents.
Multi-Source Cybersecurity Logs: An ATT&CK-Labeled Dataset and SLM Evaluation — Abstract page for arXiv paper 2606.18190: Multi-Source Cybersecurity Logs: An ATT&CK-Labeled Dataset and SLM Evaluation
Charity Majors argues AI flipped software economics: code generation is cheap, making discipline, tests, and operability—not LOC—the primary leverage for engineering teams.
Quoting Charity Majors — <blockquote cite="https://charitydotwtf.substack.com/p/ai-demands-more-engineering-discipline#footnote-2"><p>What happened in 2025 was this: <strong>the economics of code production were turned upside down</strong>. Instead of being very hard, time-consuming, and expensive to generate code, it became effectively free and instant. Lines of code went from being treasured, reused, cared for and carefully curated, to being disposable and regenerable, practically overnight...