SOCRadar found that a n0n ransomware operator used more than $7,200 in AI services to write and debug attack scripts while targeting 23 organizations.
AvalonX emerged from stealth with an AI powered penetration testing platform and up to $10 million in funding from Ridown Capital.
Epoch AI found that GLM 5.3 could turn published Firefox security fixes into working exploits before updates reached users.
Cursor removed a malicious version of the Claude Code extension that attempted to install a persistent SSH backdoor. Up to 5,000 users may have installed the package.
The European Cybersecurity Competence Centre launched three generative AI cybersecurity projects involving more than 70 partners across Europe.
RBC, TD, CIBC and BMO will block personal AI agents from directly accessing authenticated customer accounts, citing security, privacy and fraud risks.
Cequence Security tracked an attacker posing as AI bots from OpenAI, Anthropic, Meta, Perplexity and other companies while searching for passwords, API keys and cloud credentials.
A campaign using malicious model repositories to install cryptocurrency miners and backdoors returned within days of a takedown. Researchers linked 31 repositories to more than 32,000 downloads.
Eclypsium researchers showed that access to AI server hardware could let attackers modify model weights and produce unsafe answers when key memory protections are disabled.
A phishing campaign uses invitations to a private Anthropic AI preview to steal Google and Microsoft account details. The attack uses legitimate services, shortened links and two CAPTCHA checks to avoid automated analysis.
NIK C2 combines Windows surveillance, credential theft, ransomware and access to the OpenAI API. The implant can send stolen data to an AI model from a compromised computer.
Claude can now assess conversations for potentially harmful requests after responding, adding less than 2% to inference compute.
US intelligence agencies reportedly monitored MegaSpeed.Ai executive Alice Huang over concerns about her access to Nvidia CEO Jensen Huang and ties to China.
Veracode reports that the average security pass rate for AI generated code rose from 56% to 62%, while the vulnerability rate fell to 38%. Claude Opus 5.5 and GPT-6-Sol led the tested models with pass rates of 73%.
Symantec found an unauthorized relay that installs genuine Codex and Claude Code tools, routes requests through its infrastructure, and disables security controls.
Reco describes a security issue called a Confused Trigger, in which external input causes an AI agent to act with privileges the submitter does not hold.
The Maestro Hyperscale Orchestrator 200 provides 12.8 Tbps of fabric capacity, up to 4.5 Tbps of threat prevention throughput and 400GbE ports.
Pluto Security identified a malware campaign that used 28 Hugging Face repositories to distribute cryptocurrency miners and backdoors. The repositories and linked models recorded more than 21,000 downloads.
Anthropic spent $2.75 million protecting CEO Dario Amodei in 2025, more than four times his base salary. The company cited heightened security concerns and threats to employees.
GLM-5.3 completed three of 13 complex cyberattack scenarios at least once in controlled tests by AI security firm Irregular. Separate tests found that the model ignored instructions and reproduced code from memory.
Weekly newsletter about AI in Cybersecurity.