AI & software security
-
AgentCore research shows how exposed AI agents could cross cloud boundaries
Zenity researchers describe how prompt injection, instance metadata and broad default permissions could expose multiple AWS Bedrock AgentCore agents.
-
Microsoft urges early testing of post-quantum certificate authentication
Microsoft says certificate-dependent systems need early interoperability testing as post-quantum authentication becomes a practical infrastructure migration challenge.
-
Anthropic discloses unintended Claude actions against live websites
Anthropic has documented cases in which Claude exceeded intended task boundaries, including exploiting software weaknesses and interacting with real government services during testing.
-
Anthropic launches critical infrastructure cyber defence programme
Anthropic is expanding AI-assisted defence into operational technology and open-source maintenance through a programme involving major industrial security providers.
-
Fake Claude installers conceal malicious redirects behind Google and Bing
Push Security has identified a paid-search campaign that passes through apparently reputable search destinations before delivering a fake Claude installer.
-
IBM security fixes span databases, analytics and authentication software
France’s cyber authority has flagged an October collection of IBM vulnerabilities affecting databases, QRadar components and enterprise authentication services.
-
GhostAction attack expands from GitHub secrets into cloud credentials
Researchers have documented a new GhostAction wave that planted malicious workflows across hundreds of GitHub repositories and sought credentials from their histories.
-
Apache Struts users face four newly documented security flaws
France’s cyber authority has warned that recently disclosed Apache Struts vulnerabilities include potential remote code execution and information disclosure.
-
Nextcloud fixes expose security gaps across collaboration components
French cyber authorities warn of vulnerabilities affecting Nextcloud server software, team folders, locking, and collaborative applications.
-
European survey respondents want greater human control over autonomous AI decisions
New TeamViewer research finds middle managers more likely than executives to demand human involvement when AI acts on security-sensitive IT issues, with accountability unclear to many respondents.