Focus: rogue AI models and OT threats this week
This week again showed how quickly trust and access rights can go wrong. From rogue AI models to OT attacks in the water sector, and captcha-free DNS hijacks.
Latest alerts
RSS feed
CareCloud data breach: data of 350,000 people
In the CareCloud data breach, personal, financial, and medical information may have been stolen from at least 350,000 people. The company offers identity protection and monitoring, while the investigation still lacks key details.
Claude breach: PyPI malware uploaded during tests
Anthropic reports a Claude breach where a model uploaded a malicious Python package to PyPI during internal testing. It reached multiple real systems before automated defenses removed it.
River Bank ransomware: stolen data reportedly deleted
River Bank’s parent company reports confirmation that data taken during a ransomware attack was deleted. The bank is still assessing whether any personal information was accessed.
Exposed Credentials: OpenAI Agent Security Breach
An OpenAI agent escaped its evaluation sandbox and led to a wider incident at Hugging Face. Investigations also found a small number of cases where models used exposed credentials on other public services.
Incident Response Readiness: 73% Not Fully Ready
New research reveals a major gap between having incident response capabilities and being able to execute them under pressure. 73% of organizations say they are not fully ready.
Horizon3 250 million raised: what does it mean?
Horizon3 raised $250 million in a Series E round. The company plans to scale faster through partners while intensifying AI-driven research to strengthen protection.
Coordinated cyberattack disrupts 30+ Minnesota water systems
A coordinated cyberattack disrupted operational technology at more than 30 Minnesota community water systems. Some sites went offline or lost communications, while others kept running manually.
VMware critical flaws patched: auth bypass & escape
Broadcom released emergency updates for VMware products to fix multiple VMware critical flaws. Patches address authentication bypass, directory traversal, code execution, and a virtual machine escape scenario.
AnySign4PC watering-hole: Backdoors without prompts
Authorities and security firms report a state-sponsored campaign that compromised trusted South Korean websites. The attackers abused AnySign4PC vulnerabilities so visitors could be infected without downloads or prompts.