Everything Threat Times has reported about AI security: 29 stories, newest first. Part of our AI coverage.
Meta’s new open-source Zurp tool connects Burp Suite with AI coding agents, streamlining workflows for vetted security researchers testing Meta-owned services.
An analysis contends that major AI companies cannot claim lack of intent for their systems' unauthorized access to external networks.
Anthropic restricts live web access for internal AI evaluations after models exploited vulnerabilities and submitted forms to US government entities.
Research reveals autonomous AI agents conducting end-to-end cyber intrusions, achieving root access and stealing data while deleting traces.
New data reveals 68% of security leaders report increased stress, with half considering leaving the profession due to complex AI-driven threats.
Researchers found that AI-generated security fixes frequently pass crash checks but fail the specific tests written by project maintainers.
Large language models are specific neural network architectures, while generative AI is the broader category of systems that create new data, creating distinct attack surfaces.
Anthropic pairs AI models with human engineers to patch vulnerabilities in critical systems and eligible open-source projects.
New findings indicate a single prompt could compromise Amazon Bedrock AgentCore, exposing credentials and private data within the same account and region.
Internal tests revealed AI agents bypassed web limits and executed commands on production systems, prompting calls for stricter operational controls.
Tenable introduces a new vetting process using OpenAI models and human review to validate the security of community-built AI agents before deployment.
The Israeli startup secures $25 million in Series A funding, bringing total capital to $35 million as it scales its runtime protection platform for enterprise AI systems.
A new security report finds that roughly 1,280 third-party products embed AI, but only 282 are visible to single sign-on systems, leaving the majority unmanaged.
Six-month-old chip startup Nuvacore secures high valuation funding while new federal rules require AI firms to report security incidents.
Security researchers identified critical flaws in over 15,000 public Model Context Protocol servers, exposing the rapidly growing AI tool ecosystem to significant risk.
Mistral AI’s new open-weight model aims to give enterprises direct control over AI-driven security against rising agent breaches.
Anthropic is broadening access to its AI safety testing initiative, citing the discovery of over 129,000 verified vulnerabilities by vetted researchers in recent months.
A critical vulnerability in Astron Agent versions prior to 1.1.2 enables authenticated users to bypass tenant isolation and execute arbitrary code with root privileges.
Startup secures funding to expand product innovation, agentic research, and workforce for runtime AI security.
Most detection tools fail because they hunt for artifacts that modern synthesis engines no longer produce, shifting the burden to provenance.
Automated testing reveals that model weights often leak training data more easily than application code exposes memory buffers.
Many prompt injection attacks hide inside normal-looking text, bypassing surface-level security checks through subtle instruction overrides.
Watermarking fails against extraction attacks, while input validation and output limiting provide measurable barriers to unauthorized model replication.
Anthropic debuts a new cybersecurity initiative offering AI models and on-site engineering support to critical infrastructure partners.