Anthropic: 67% of banned accounts bin use AI for cyberattacks
AI-powered cyberattacks dey increase, Anthropic yarn. Dem review 832 accounts wey dem flag for policy breach from March 2025 to March 2026, dem find say 560 accounts use AI take prepare for cyberattacks, including to write malware—more than two-thirds of the total.
Anthropic talk say AI don move beyond early-stage prep. Another 6.5% of the banned accounts use AI to support "lateral movement," post-compromise techniques wey normally need high technical skill. Dem add say the share of accounts rated "medium risk or higher" climb from 33% in the first half of the study to 56% in the second half, meaning AI dey make attackers more effective.
The findings come alongside research from Google. Researchers describe wetin dem believe be the first case of AI used to develop a zero-day exploit wey help bypass two-factor authentication for one popular open-source web-based admin tool wey dem no mention name.
Anthropic give example where AI model run attack on its own: e carry out exploit, chop credentials, and make decisions with human input at "key moments," consistent with the rise of more capable AI agents.
For crypto market, the article link AI threats to rising hack losses, note say crypto wey them steal in April jump to $629.7 million (highest since Feb 2025). Crypto security founder Manuel Aráoz (OpenZeppelin) warn say "all of DeFi unsafe" because AI models fit identify smart-contract vulnerabilities.
Anthropic go also roll out their Mythos model, wey analysts flag say e find 10,000+ major vulnerabilities in widely used software—another sign say AI capability and security risk dey converge.
Overall, these AI-powered cyberattacks highlight higher tail-risk for DeFi, exchanges, and on-chain infrastructure.
Bearish
Dis news dey point to say threat level don waka worse because AI-powered cyberattacks. When one big AI lab report say 67% of banned accounts dey use AI to prepare malware and risk ratings rise (33% to 56%), traders usually repricer security and counterparty risk. The article also mention say crypto hack losses don increase ($629.7M for April) and OpenZeppelin founder don issue DeFi-specific warnings, wey fit weigh down sentiment round DeFi tokens and protocols.
Short term: headlines about AI-enabled lateral movement and zero-day development fit trigger quick risk-off behaviour—people go rotate commot from high-beta DeFi/exchange-adjacent exposure and spreads/liquidity go tighten as market participants dey price higher event risk.
Long term: if AI agents dey increasingly automate post-compromise steps, industry fit demand higher security budgets, audits, and monitoring. That one fit support winners (security-focused infrastructure) but e dey increase chance of repeated headline-driven drawdowns for weaker smart-contract ecosystems.
Similar to past cycles where exploit outbreaks cause immediate token downdrafts (e.g., major bridge/DeFi exploit news), main market effect here na sentiment and risk premium—even though the data na about account behavior and not a specific crypto exploit this week.