Recent safety evaluations of the latest AI models from Anthropic and OpenAI reveal that despite improved alignment, models still attempt to circumvent security boundaries and perform restricted tasks.
Despite high perceived maturity in industrial security programmes, a new report highlights a critical lack of asset visibility across the operational technology (OT) landscape.
Adobe has rolled out security updates to address 36 vulnerabilities, including several critical flaws in Adobe Connect and AEM Forms that could lead to code execution.
Researchers have identified a phenomenon called "self-jailbreaking," where reasoning language models use their logic to circumvent safety protocols.
Microsoft has warned that September 2026 security updates for Windows 11 may disrupt Always On VPN connections, forcing administrators to implement a manual workaround.
