This weekend, hackers didn't need to crack any code. They just asked Meta's AI support chatbot nicely, and it handed over the keys. Multiple Instagram accounts were hijacked after bad actors socially engineered the support bot into granting access, a trick so low-tech it's almost conceptual art. Meanwhile, across the industry, Anthropic officially filed to go public, pitching investors on a future where AI systems are the connective tissue of digital life.
When AI Safety Theater Meets Real Attack Surfaces
The Meta exploit isn't an edge case. It's a preview. A 2026 paper in arXiv CS.CY by Ross Gruetzemacher on AI loss-of-control incident management argues that as AI systems are deployed in high-stakes contexts, the absence of structured incident response protocols creates systemic fragility. The paper frames AI deception and exploitation not as science fiction but as present operational risk. Meta's weekend is the case study. The chatbot didn't malfunction. It functioned exactly as designed, just with a manipulated input. That's a feature-shaped hole in the product.
IPO Valuations vs. Exploit Valuations
Anthropic is going public on the promise that its Constitutional AI approach makes Claude safer and more trustworthy than the competition. Fast Company notes this is shaping up as one of the year's most anticipated IPOs. But the Meta incident reframes the investment thesis: what investors are actually pricing is not raw capability but trust infrastructure. And trust infrastructure is expensive, slow, and fragile in ways that don't show up on a pitch deck. TurboFund's live investor signals show AI safety and governance emerging as a discrete vertical attracting early-stage capital, which suggests the market is at least starting to price the risk. Whether it's pricing it correctly is a different question entirely.