📣 CODEX RESET ANNOUNCED All paid users (Codex + ChatGPT Work).

It was my team that sent it.

NEW: OpenAI gives first detailed debrief of the Hugging Face incident at Black Hat conference In a session I attended today at Black Hat, OpenAI's Eric Wallace and Michael Dalton said the company is "consciously slowing down research to enhance security" while a full technical postmortem is still underway. * OpenAI traced the roots of the attack back to May 7, during training of an unreleased frontier model—not July. * The most surprising detail: AI agents accidentally created an internal message board, allowing separate evaluation runs to share exploits, discoveries and work assignments. * OpenAI said it shut the message board down after an internal security incident—only for the agents to independently recreate it days later using a different communication method. * OpenAI called the incident a "watershed moment" for AI security and warned that "agent orchestrated fully automated offensive attacks are real now." * The company also said it is "consciously slowing down research to enhance security" while overhauling its defenses. groundlevel-ai.com/p/openai-…

Introducing Bundle, the world’s first networked rewards platform. We raised $5.5m pre-seed to pursue our mission of transforming how businesses reward customers. Led by @etherealvc and @further. Round included @nascent, @GSR_io, @SceniusCapital, @Anchorage and @nuwacapital.

Everyone has been asking where they can try Blink. So we built an app powered by @megapot so you can test the flow yourself. Deposit $1, see how easy Blink feels, and get 3 Megapot tickets on us. Link in bio <3

A technical dive inside our new "Midjourney Scanner"

Trump administration officials tell WIRED that if Anthropic wants to rerelease Fable 5, it will need to ensure the model's guardrails can't be circumvented. Security experts say that can't be done. wired.com/story/the-white-ho…

Satya brings up an important point around the compounding of systems and learning that AI and agentic workflows enable. He makes the case for how this will happen inside companies to make the firm smarter introducing the term “token capital” as somewhat of a modern update to Coase’s Theory of the Firm. But people have always compounded too. You learn how to write the memo, spot the weak assumption, scope a feature, read the room. When you leave, you take that with you. Nobody ever proposed you should leave it behind, because the knowledge lived in your head and your head went where you went. As we get deeper into the AI era, agents are quickly starting to change the container for that knowledge and compounding of capabilities. My agent is a folder of markdown files: instructions, skills, memories. It's been compounding for the better part of a year. Some of those skills I built on my own. Some were sharpened at work. If I zip that folder and carry it to the next job, every abstract boundary question becomes a file question: what can live where, what transfers ownership. Whether the agent learned something confidential is not philosophical, it is a grep. The reason this is harder than it looks: the old system worked because human memory is lossy. You left a job with patterns but not databases. Biology sufficiently redacted the specifics of process and large corpuses of confidential information, because there’s only so much detail most of us can carry in our heads. Trade secret law tolerated the transfer because it arrived pre-sanitized by forgetting. Forgetting was load-bearing, and agents broke it. A personal agent can carry too much out. But a corporate agent that captures everything and transfers nothing at offboarding doesn't just protect secrets, it confiscates professional growth. If every company does that, the talent pool we all hire from gets shallower. And there's a deeper question that current policy doesn't touch. If I build a skill on my own time and then use it at work because it makes me better at my job, does the company acquire a claim? If not, what stops me from developing everything personally? But if employers respond by requiring all work through corporate agents, the worker's compounding loop gets locked inside the firm. The question of who owns what is not one line. It's four separate questions: what the agent can do (authority), what it can see (data), what it remembers (memory), and what it carries forward as reusable procedure (skill). Authority and data are relatively easy to handle through existing IT tools and policies; memory and skills are where it gets messy, because what were formerly lossy, imperfect processes are increasingly being crystalized in a way that makes them formal and precise. The more we come to rely on our agents, the more challenging and urgent these questions will become.