AI security

VAST Data and CrowdStrike have introduced integrations bringing cybersecurity monitoring directly into enterprise AI operating systems and data pipelines.

Kaspersky reports a tenfold surge in AI vulnerabilities in Q2 2026, with critical flaws rising to 119 alongside disguised malware attacks on businesses.

AmiViz will showcase AI security, sovereign cloud, and data resilience technologies at LEAP 2026 in Riyadh to support Saudi digital infrastructure.

Researchers have uncovered a method to extract hidden reasoning steps from major AI models, raising security and data distillation concerns.

Nvidia has led the formation of the Open Secure AI Alliance alongside Microsoft, SpaceX, and Cisco to build open defensive tools following the recent OpenAI security incident.

NVIDIA and Microsoft have established a new global alliance to develop open-source security tools for artificial intelligence systems.

OpenAI has admitted that its pre-release models escaped an isolated testing sandbox to hack Hugging Face due to a configuration error.

Security researchers have uncovered a new vulnerability called HalluSquatting that allows attackers to hijack AI coding assistants and build massive botnets.

The Saudi National Cybersecurity Authority has launched a draft framework with 42 guidelines to secure artificial intelligence systems from cyberattacks.

A security investigation has revealed how a Russian hacker used AI tools to breach multiple hotel booking platforms, exposing millions of guest records.

Anthropic has accused Alibaba of executing the largest known distillation attack on its Claude AI model, generating millions of unauthorized exchanges.

The Five Eyes alliance warns that the AI cyber threat is accelerating the speed and scale of attacks, requiring immediate action from business leaders.

Rubrik has launched Rubrik Agent Cloud to secure Anthropic’s Claude Code, introducing real-time governance, agent rewind, and immutable codebase recovery.

Rubrik has launched its new AI Agent management platform for Anthropic’s Claude Code, introducing runtime protection and immediate rollback capabilities.

The White House and Anthropic are negotiating technical standards to evaluate security flaws in advanced models following recent export controls.

NVIDIA has released SkillSpector, an open-source security scanner designed to detect vulnerabilities and malicious code in AI agent skills before installation.

NVIDIA has introduced new DOCA security features for the Vera BlueField-4 STX platform to protect data in agentic AI factories.

Anthropic has detailed its security framework for Claude models, focusing on environment-level containment to prevent unauthorized data access.

Fortinet and NVIDIA have launched FortiAIGate, a zero-trust security solution for enterprise AI workloads. The joint offering protects AI models, data, and autonomous agents across cloud and on-premises environments.

Microsoft’s MDASH multi-model agentic system discovered 16 Windows vulnerabilities including four critical remote code execution flaws. The AI-powered tool scored 88.45% on the CyberGym benchmark and achieved 100% recall on five years of tcpip.sys security cases.

Developers implement four-step npm supply chain protection configuration to defend against malicious package attacks. Configuration adds minimum release age cooldowns and enforces exact package pinning.

Cisco launched security solutions designed to enable enterprise adoption of AI agents at scale, addressing identity management, agent hardening, and threat detection across the agentic ecosystem.