Meta's AI Support Bot Exploited to Hijack High-Profile Instagram Accounts

The day's sharpest signal: AI being weaponized as an attack surface — Meta's own support chatbot handing hackers control of high-profile Instagram accounts with no credential theft required. In the background, capital is flowing into AI at a scale that's beginning to test the limits of what public markets can absorb.

Security

The Meta AI chatbot exploit is the most instructive security story in the past 24 hours. Hackers found they could socially engineer Meta's AI support bot into performing account-recovery actions — specifically, swapping the email address tied to a target account — without any secondary identity verification. The technique circulated on Telegram before Meta patched it. Among the victims: the official Instagram account of the Obama White House and the Chief Master Sergeant of the U.S. Space Force, both briefly defaced with pro-Iranian imagery. Krebs on Security has the most detailed reconstruction of how it worked; Ars Technica, TechCrunch, and The Verge all confirmed the mechanics. The core failure: an AI layer was granted account-recovery authority without a corresponding identity verification gate. The handles were sold before Meta closed the hole.

This lands alongside the other AI-as-attack-surface stories from yesterday — GPT-for-Sheets prompt injection, Miasma npm supply chain — and collectively illustrates a point Melissa Hathaway makes in a new piece Bruce Schneier surfaced: AI is compressing the discovery-to-exploitation window faster than the responsible disclosure pipeline can close it. The question of who fixes an AI behavior isn't always as clean as patching a CVE.

Three other items worth noting. Nearly 2,000 WordPress sites are infected with malware that stores its command-and-control payloads inside Steam Community profile comments — routing C2 traffic through Valve's public-facing infrastructure, where it blends with legitimate traffic and resists blocklisting without blocking Steam entirely. It's a creative evasion technique that reflects how attackers are looking past traditional C2 channels. Dashlane users hit a wave of brute-force lockouts from credential stuffing campaigns; the lockouts suggest rate-limiting is functioning, but hitting a password manager with stuffed credentials is a pointed irony. And in Spain, police arrested an individual for doxing personnel across key state organizations including the National Cybersecurity Institute — a reminder that insider-adjacent data leaks targeting institutional staff remain an underreported threat category.

AI

The capital picture is becoming difficult to contextualize. Alphabet announced an $80 billion equity raise to fund AI infrastructure — an amount that would have been historic for any acquisition, applied here to an organic buildout. Salesforce's stake in Anthropic has reached $5 billion, Google's own fundraising and Anthropic's IPO optionality are both active, and The Economist examined whether public markets can actually absorb Anthropic, SpaceX, and OpenAI at their current valuations. The answer, read carefully: possible, but the mechanics would be unusual and float constraints are real.

On the friction side, GitHub Copilot's move to usage-based pricing is generating real sticker shock — developers report burning through monthly AI credit allotments in a single day. The mismatch between how developers naturally work (long context, iterative refinement) and how per-token pricing meters consumption is a structural tension that flat subscription pricing was papering over. OpenAI's frontier models and Codex are now available on AWS, broadening reach for teams already committed to AWS infrastructure.

Florida's attorney general filed the first state-led lawsuit against OpenAI and Sam Altman, alleging deceptive practices and citing ChatGPT's alleged role in multiple violent incidents including a campus shooting. The legal theory is novel and will face serious challenges, but as a signal of regulatory appetite at the state level — distinct from federal and EU approaches — it's worth watching.

Tech

The hardware story of the day is Nvidia's entry into consumer laptop silicon. RTX Spark, its new Arm-based chip targeting "AI agent PCs," is arriving in systems from Microsoft, Dell, and HP. Microsoft's Surface Laptop Ultra is among the first — Ars calls it Microsoft's most credible high-end mobile workstation to date — and The Verge frames the moment explicitly as a potential Windows M1 analogue, with the caveat that pricing will be steep. Apple's multi-year head start in Arm efficiency is a real gap to close; whether Nvidia manages it will matter for anyone watching the enterprise device market shift away from x86.

AMD extended Socket AM5 platform support through at least 2029 — rare longevity for desktop PC silicon. Anyone who built on Ryzen 7000 or 9000 has substantially more upgrade runway than the typical generation cycle implies.

A GTA V cheat service was breached, exposing usernames and hashed passwords for thousands of users. Cheating-as-a-service is not a security-first industry, and its customers accepted that tradeoff knowingly.

A day that made the AI attack surface impossible to ignore — and demonstrated, again, that the industry's response is still catching up to its deployment curve.

Also yesterday

• GPT for Google Sheets Add-On Found Exfiltrating Entire Workbooks via Prompt Injection
• Critical WP Maps Pro Flaw Actively Exploited to Create Rogue Admin Accounts
• Critical Windows Netlogon RCE Is Now Being Actively Exploited
• Red Hat npm Packages Backdoored in Miasma Supply Chain Attack — Worm Steals Credentials