OpenAI’s Codex helps discover HTTP/2 Bomb DoS attack that can nuke over 30GB of RAM within seconds, knocking web servers offline before they can react
THE SO WHAT
AI just helped find an HTTP/2 attack that can blow through 30GB of RAM in seconds—offense is now using the same tooling as defense. Treat advanced code models as part of your red team this quarter, and assume that any public protocol surface is being fuzzed by AI at machine speed already.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIAI Scientist Bengio: Building Systems We Don't Know How to Control
When Yoshua Bengio says we don’t have control of AI agents, he’s giving operators cover to slow deployment and demand governance budgets. If your board hasn’t asked for an agent risk register and escalation plan yet, this is the interview that will trigger it.
Applied AIUpwind, the next-gen wiz, now secures every corner of the AI stack
“Security for AI” as a thesis—spanning models, data, and agents—means AI risk is getting its own category, not just a feature in cloud security. If you’re shipping agentic workflows into production, expect your CISO to demand a dedicated AI security vendor on the architecture diagram this quarter.
Applied AIPoke, which lets users access AI agents via text message, becomes the first AI agent approved for Apple's Messages for Business platform (Sarah Perez/TechCrunch)
An AI agent inside Apple’s Messages for Business turns every customer text thread into an automation surface—no app, no login, just SMS. If you run support or commerce, you now have to assume your customers will expect agentic resolution inside the default messaging channel, not your bespoke UX.
Applied AIAnthropic says 80% of its new production code is now authored by Claude — how your enterprise can keep up
If 80% of Anthropic’s merged production code in May came from Claude, the constraint on software velocity is now review bandwidth and test coverage, not headcount. If your SDLC, QA, and compliance stack aren’t re-architected around AI-first code generation this year, your roadmap assumptions are already stale.