Cybercriminals are increasingly attacking AI-based protection mechanisms directly, while AI-powered website builders and OAuth-based consent phishing open new attack surfaces.
AI risks in enterprises concentrate on five percent power users and private consumer AI accounts, while enterprise solutions provide significantly better governance.
Just-In-Time Access replaces permanent access with automatically expiring time-limited permissions and reduces the exploitation window for compromised cloud identities from months to hours.
A self-learning framework for code-repair agents leverages their solution traces directly to generate targeted training tasks, achieving higher accuracy than previous approaches.
Meta installed facial recognition with three AI models and local biometric storage in the Ray-Ban glasses app and disabled it via server command—without informing users.
While China seeks access to US cyber AI models, the US industry is racing to deploy these models for defensive measures quickly enough – but time is running short.
AI outputs are economically valuable only when humans assess their correctness and relevance to the business context, rather than adopting them blindly.
The use of AI for mass production of content causes AI systems to increasingly learn from other AI-generated content, allowing errors and biases to accumulate rather than be corrected.
Attackers can inject malicious commands into messenger messages through fake context alignment, which Gemini processes undetected and uses to control authorized devices or misuse data.
