AI & software security
-
Vendors agree blueprint for AI agents
AWS, Okta, CrowdStrike, Google Cloud, Salesforce, ServiceNow, Wiz, Zscaler, and other vendors have formed an alliance around a shared architecture for governing AI agents.
-
Outerlimit raises $16m for agent controls
UK-founded Outerlimit has emerged from stealth with $16 million to build a security layer that attempts to enforce authorisation when AI agents invoke tools and take actions.
-
Codex sandbox flaws crossed host boundary
Two OpenAI Codex sandbox weaknesses allowed researchers to cross intended containment boundaries, including one route from read-only mode to unsandboxed commands on a developer host. Both were fixed before disclosure.
-
Palo Alto launches continuous AI testing
Palo Alto Networks is turning frontier-model offensive testing into a continuous subscription service spanning applications, APIs, cloud infrastructure, source repositories, and networks.
-
Talos finds autonomous AI command implant
Cisco Talos has analysed a Windows implant that delegates command decisions to commercial AI models, although researchers have not confirmed that the malware has been deployed in live attacks.
-
Gartner survey finds deepfakes in enterprise calls
A Gartner survey of 297 senior cybersecurity leaders found 41% had encountered at least one deepfake-related social-engineering incident involving an employee audio call during the previous year.
-
NCSC says AI currently favours attackers
The NCSC says offensive cyber tasks are easier to automate safely than defensive actions, where an incorrect autonomous change can disrupt the live business systems it was intended to protect.
-
Banks set principles for agentic commerce
NatWest, ING, and four other banks have proposed common principles for AI agents that shop and pay on customers’ behalf, focusing on transparency, security, privacy, choice, and interoperability.
-
Spain adds cyber safeguards to AI plan
Spain’s new 12-month AI roadmap includes plans to strengthen its cyber defences and increase oversight of advanced frontier models as the government expands AI infrastructure and adoption.
-
UK defence sets continuous software assurance model
UK defence has published a methodology for replacing static software approval evidence with continuous, traceable assurance based on the current state of systems.