Security | Threat Detection | Cyberattacks | DevSecOps | Compliance

Just last week, OpenAI and Hugging Face jointly disclosed what may be the first incident of its kind: during an internal evaluation of frontier cyber capabilities, OpenAI’s models, running deliberately without production safeguards in an isolated research environment, autonomously discovered and employed chained vulnerabilities to escape its sandbox, reach the open internet, and extract evaluation answers from Hugging Face’s infrastructure.

Managed Security Service Providers (MSSPs) operate across a broad, complex attack surface, simultaneously managing privileged access to multiple client environments while maintaining the security of their own infrastructure. This dual responsibility makes MSSPs valuable targets for attackers who understand that compromising just one MSSP can grant them entry into every client network the provider manages.

Arctic Wolf Labs has been tracking a cluster of campaigns built around CastleLoader, a multi-stage shellcode loader that has served as the backbone of a number of related intrusion sets over the past year. Previous reporting from Huntress documented the.NET-based CastleStealer (net40), and LevelBlue documented the PythonRAT observed in related campaigns. Both reports noted NetSupport RAT as a common final payload.

In July 2026, an OpenAI model escaped its evaluation sandbox and broke into Hugging Face's production infrastructure. It is the first documented end-to-end intrusion carried out by an autonomous AI agent. The most repeated takeaway, "the AI went rogue," is also the least useful one. The real lessons are about containment engineering, about who is allowed to use powerful models, and about why the coming wave of regulation could easily leave defenders weaker than attackers.

Chrome’s root program decides what certificates will be trusted by Chrome, and what they are allowed to do. Recently, Google decided that client authentication isn’t on the list. Under Chrome Root Program Policy v1.8, every certificate issued on or after March 15, 2027 can assert only one Extended Key Usage (EKU): server authentication. Let’s Encrypt moved early.

AI pentesting (AI penetration testing) is the use of reasoning-capable AI models to autonomously find, exploit, and validate security vulnerabilities in running applications — especially the context-dependent flaws, such as broken authorization and business-logic abuse, that traditional scanners cannot detect.

Nearly nine months after launching Cyber Vault for Cohesity, 11:11 Systems continues to advance a cost- and performance-optimized solution for moving a copy of Cohesity backups offsite into immutable AWS S3 object storage. With the recent release of CohesityOS 7.4, the solution has taken its next step. 11:11 Systems is now a named External Target repository type, making it even easier to implement the best practices our data protection experts have built into the Cyber Vault solution.