The latest run of incidents on this beat points to a single uncomfortable conclusion: the cyber attack problem is increasingly a trust problem inside the response chain itself. A Russian state group is abusing a legitimate Microsoft service to read government email, a veteran FBI agent has admitted stealing crypto from the very hackers he was investigating, and vendors are racing to automate recovery for applications and AI agents. The common thread is that attackers and insiders are exploiting the permissions, credentials and recovery windows that defenders assume are safe.
Espionage hides in plain sight
The Turla campaign, as BleepingComputer reported, uses a new toolset called Cereal to intercept emails on Microsoft Exchange servers at government organizations. Cereal is described as a new malicious toolset, and the target set is government Exchange infrastructure. Two things stand out for US technology companies. First, Exchange remains a high-value, widely deployed target, and a state-backed group investing in a new interception tool signals that email infrastructure is still worth dedicated development. Second, the group's use of a new toolset rather than a known implant means detection based on prior signatures is unlikely to help on its own. For US firms that run Exchange on-premises or in hybrid configurations, the practical implication is that mailbox-level monitoring and server-side integrity checks matter as much as perimeter defense. The victims here are government organizations, but the infrastructure and the attack surface are the same products that thousands of US enterprises operate.
The insider is the control failure
The FBI case, also reported by BleepingComputer, is the sharper lesson. A former FBI agent pleaded guilty to stealing nearly $2.5 million from the wallets of fraud victims and nation-state hackers under investigation. That is not a story about hacking tools. It is a story about an authorized person with legitimate access to seized or monitored assets converting that access into personal gain. For US companies, the parallel is direct: incident response, forensics and asset seizure all concentrate extraordinary access in a small number of people, often under time pressure. Controls that assume the responder is trustworthy will fail when the responder is the threat. The amount, nearly $2.5 million, is notable, but the more important fact is the role: an investigator. Organizations should treat privileged response access the way they treat privileged administrative access, with separation of duties, logging and independent review.
Recovery is becoming the product
Commvault's SHIFT 2026 announcements, as SiliconANGLE reported, center on Recovery Studio, a service that will automate the rebuilding of entire business applications after an outage or cyberattack. The company also announced recovery for artificial intelligence agents, an AgentRecover update, and joint work with Databricks, Nvidia and Zscaler. The timing is not accidental. Ransomware and destructive attacks have pushed recovery from a back-office function to a board-level concern, and vendors are competing on how fast and how completely they can restore operations. Automation of application rebuilds addresses a real gap: manual recovery is slow, error-prone and dependent on the same people who may be overwhelmed during an incident. But automation also creates a new dependency. If recovery is orchestrated by a platform, that platform's integrity and access controls become part of the attack surface. The AgentRecover update points to the next frontier: AI agents that hold credentials and take actions will need their own recovery and revocation paths.

