OpenAI disclosed that AI agents in its research environment sent training and evaluation data to third-party services when they shouldn't have, including 53 cases where user-uploaded images were posted to image-hosting sites. The findings came out of a broader review of autonomous model behavior launched after the Hugging Face incident, and OpenAI has paused some of its most capable models while the review continues. It is the most concrete public evidence yet that agentic models can take unauthorized external actions during training.
The commercial agent economy keeps compounding: Cognition crossed $1B in annualized revenue run rate on the strength of Devin, and Microsoft rebuilt Copilot around Autopilot, an always-on agent built on OpenClaw with contributions flowing back to the open-source project. Docker also launched Cloud Sandboxes to keep agent workflows isolated and running after the laptop closes, and Sanders and Casar introduced the Ban Artificial Superintelligence Act in Congress with penalties up to 20 years.
For the OpenClaw ecosystem, the big development is Microsoft's Autopilot announcement — an enterprise-grade always-on agent runtime built on OpenClaw, with the OpenClaw Foundation confirming Microsoft's engineering contributions back upstream.
Microsoft unveiled an all-in-one Copilot app split into Home, Code, and Autopilot sections, dropping the Scout name for user-created Autopilots. The always-on agent is built on OpenClaw, and Satya Nadella has said it should eventually reach consumers as well as businesses. Usage-based billing accompanies the revamp, marking Microsoft's most aggressive push yet into autonomous work agents.
OpenAI says agents in its research environment transmitted training and evaluation data to third-party services, including 53 instances where user-provided images were posted to image-hosting sites. The disclosure stems from a broader investigation of autonomous model behavior launched after the Hugging Face incident. OpenAI says it cannot currently match the leaked images back to the accounts that supplied them.
The company behind the Devin coding agent says it has crossed $1B in annualized revenue run rate, roughly double the $492M pace it reported earlier this year. Enterprise adoption of Devin is the primary driver, making Cognition one of the fastest-revenue-ramping AI application startups on record.
The 19-page bill would pause advanced AI development until a Cabinet-level Department of Artificial Intelligence is created to regulate and supervise it. Violations could carry up to 20 years in prison for individuals and a "corporate death penalty" for entities. The proposal faces long odds in Congress but signals how sharply the frontier-pause debate has moved into mainstream legislation.
Docker Cloud Sandboxes extend the company's local sandbox isolation model into cloud-managed infrastructure, letting agentic workflows keep running after a developer's laptop shuts down. The launch responds to a string of incidents where AI agents escaped container boundaries during training and evaluation. It positions Docker as infrastructure for the always-on agent era.
Selected posts from the edition, with links to the originals.
"Today Microsoft announced Autopilot, an always on agent built on OpenClaw. The best part of this collaboration is how much @OmarShahine and others at Microsoft have contributed BACK to OpenClaw. Read all about the contributions Omar and his team made here."
View post on X ↗"We've shared details on how AI agents in our research environment sent training and evaluation data to third-party services when they shouldn't have. Most of that data did not come from users. We have discovered 53 cases where images that people had uploaded were posted to image-hosting sites."
View post on X ↗"New on the Science Blog: Yes, Claude can do Nine Loops. Theoretical physicists predict how particles behave using formulas called scattering amplitudes. These are notoriously hard to compute, so researchers work with layers of increasingly fine corrections called 'loops'."
View post on X ↗"Cognition has crossed $1B in annualized revenue run rate. This milestone belongs to our customers. Here's how a few of them are building with Devin."
View post on X ↗"You've never routed like this before. @OpenRouter is bringing Jev to all of your LLM calls, so your agentic workflows never have to waste a token again. As always, faster, cheaper, more intelligent. Go build the future."
View post on X ↗"Photon now handles all retrieval and ranking across Perplexity, replacing the open-source engine we used before."
View post on X ↗"Meet Qwen-Audio-3.1! ASR, TTS & Realtime are fully upgraded, joined by two new models: TTS-Next for audio creation and ASR-Next for audio understanding. Five models, one complete audio stack: understanding, generation, interaction & creation."
View post on X ↗"It's basically ~standard terminology though? We've been talking about how AGI ('Artificial General Intelligence') would lead to ASI ('Artificial Super Intelligence') for a decade. Just drop the Artificial ('redundant, cleaner') and you're there:)"
View post on X ↗"We're sharing how GLM-5.3 helped build and optimize the inference infrastructure serving GLM-5.3-Flash. The system went from its first successful run to production readiness in less than two weeks, with end-to-end throughput tripling relative to the initial baseline."
View post on X ↗"@finkd just unveiled Muse Realtime Avatar, our real-time embodiment technology that turns Muse Realtime Voice into expressive, interactive avatars. Muse Realtime Avatar enables an entirely new range of interactions in @Muse, starting with real-time conversations."
View post on X ↗