Britain steps up its defences against AI risks
THE SO WHAT
Finding 407 vulnerabilities across nine agencies for £13,000 in tokens is a proof point that AI-augmented red teaming is cost-effective at scale. CISOs should assume regulators will adopt similar tools—and preemptively use AI to probe their own estates before someone else does.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIMeta says Wonder includes Active Memory and Active Focus, two features designed to help AI systems perform more useful work over extended periods
Active Memory and Active Focus are about turning stateless chatbots into long-running operators that remember, prioritize, and stay on task. If you’re designing workflows around agents, assume Meta is optimizing for continuous, multi-session work rather than one-off prompts.
Applied AIMeta says it plans to power its Muse AI agent with Wonder, starting with subscribers, and will develop specialized models it "can bring to everyone using Muse"
Tying Wonder first to Muse subscribers makes the AI frontier a monetization lever, not just a research milestone. If you’re building on Meta’s stack, expect a tiered capability model where the best agent behaviors live behind subscription or enterprise SKUs.
Applied AIAn interview with Mark Zuckerberg, who says Meta has caught up with rivals at the frontier of AI with its Wonder model and rejects calls for a slowdown
Meta framing Wonder as “caught up” and explicitly rejecting a slowdown is a commitment to pace and spend, not just a model drop. For operators, that means Meta will likely keep pushing open releases and aggressive productization—plan for a fast-moving baseline rather than a stable platform.
Applied AIMuse Wonder Puts Meta Back at the AI Frontier
Meta is pairing raw model quality with guardrail features specifically aimed at agent reliability. If you’re evaluating front-end assistants, the question shifts from “who’s smartest?” to “whose agents can run unattended without going off the rails.”