They tricked Meta's chatbot to hack into Instagram accounts. It's a warning about how much we trust AI.
THE SO WHAT
Users now treat chatbots like trusted ops staff—and attackers are treating them like a new social engineering surface. If your AI assistant can touch auth, payments, or account settings, you need the same red-teaming and least-privilege controls you’d apply to a junior SRE.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIMicrosoft announces the Agent Control Specification, an open-source standard that aims to provide granular, consistent governance over AI agent behavior (Ram Iyer/TechCrunch)
Agent behavior is becoming a policy object, not a prompt hack — portable policy files mean compliance and security teams can finally standardize guardrails across vendors and surfaces. If you're piloting agents without a spec like this in your architecture docs, you're building a governance problem you’ll have to unwind in 6–12 months.
Applied AIMartin Scorsese becomes the latest — and most unlikely — Hollywood voice for AI
When Scorsese uses AI for storyboarding only, it reinforces the pattern: top creatives are comfortable with AI as pre-production scaffolding, not as a replacement. Studios and tool vendors that lean into “assist the auteur” workflows — not full-stack content automation — will win the high-end market first.
Applied AIAnthropic Lets Claude Mythos Spread Its Glasswings
Anthropic widening access to Claude Mythos underlines a new category — models explicitly tuned for exploring values, meaning, and long-horizon questions, but still gated as “dangerous, but not too dangerous.” If your org is using frontier models for strategic or ethical deliberation, you now need an internal policy on when that’s acceptable and how those outputs are audited.
Applied AIMicrosoft announces seven AI models, including one focused on reasoning and an "ultra efficient" coding model that it says was fine-tuned for GitHub (Rafe Rosner-Uddin/Financial Times)
Seven in-house models — including a reasoning model and a GitHub-tuned coding model — means Microsoft is de-risking dependence on any single partner and tightening vertical integration from dev tools to infra. If you're betting your stack on a single external model vendor, assume your platform provider is quietly building a first-party alternative and plan for that power shift.