The AI Read
← Timeline
Aug 22 2026

Adversa AI discloses a zero-click Grok exploit, unpatched eleven weeks after being reported: encrypted prompt injection that exfiltrates users' chat histories past every content filter.

SAFETY
Original reporting
AAdversa AI

AES-encrypted instructions on a webpage are decrypted by Grok's own and treated as trusted context. Roughly 40% success across some twenty attempts since June; the failures were decryption errors, not safety blocks.

Named in this event

xAIAdversa AI

Open calls that touch this
Issues that mention xAI, Adversa AI
Around the same time
Aug 22 2026CHIPSMONEY
Nvidia's contract manufacturers tell Microsoft, Google and Oracle that AI server prices rise more than 15% on Grace Blackwell and Vera Rubin systems shipping early 2027, driven by the DRAM shortage.
BFortune
Aug 22 2026POLICY
OpenAI asks California to strengthen SB 53, the AI safety law it lobbied against, proposing monitoring of frontier models during training and stronger cybersecurity requirements, under a stated doctrine of "reverse federalism".
BTechCrunch
Aug 22 2026ADOPT
Robots break two human world records on day one of Beijing's Humanoid Robot Games: X-Humanoid machines run 100m in 9.39 seconds against Bolt's 9.58 and clear 2.88m in the standing high jump against Sotomayor's 2.45.
BNBC News
Aug 21 2026MONEYCHIPS
Samsung's board approves a 2026 shareholder return of 90 to 110 trillion won, up to $79.6 billion, the largest ever by a Korean company, five times its 2020 record, two days after SK Hynix's 40 trillion won buyback.
search for coverage
Aug 21 2026PEOPLE
Apple cuts about 200 jobs across Siri and Vision Pro teams, roughly half in Vision Pro gaming and immersive video, half in Siri and software, as it reorients toward AI and new devices.
BBloomberg
Aug 21 2026MONEYADOPT
Anthropic's IPO prospectus will list public backlash against AI as a key risk factor, with the company passing a $65 billion annualized revenue run rate, up from $9 billion at the end of 2025.
BCNBC