The three stories logged on this beat share a single thread: attackers are increasingly working through the trusted infrastructure that organizations depend on rather than battering at the organizations themselves. SafePay's reach, the theft of UpdraftCentral connection keys, and the criminal use of agentic AI all describe attacks that borrow someone else's access, scale, or credibility. For US technology companies, that shifts the defensive problem from hardening the perimeter to auditing every dependency that can act on their behalf.
Scale Through Other People's Systems
CISA's assessment, as BleepingComputer reported, is that the SafePay ransomware gang has breached more than 600 critical infrastructure organizations worldwide since 2024. The figure matters less as a body count than as a description of method. Ransomware crews have long preferred the easiest available entry, but a campaign of that breadth against critical infrastructure implies industrialised targeting rather than opportunistic break-ins. The victims are the visible end of the chain; the leverage sits in whatever access, tooling, or insiders let the group repeat the operation hundreds of times.
The lesson for US operators is that criticality does not confer immunity. Organizations in energy, water, transport, and healthcare are attractive precisely because their downtime is expensive and their supply chains are interdependent. A single compromised vendor or shared service can put many of those organizations in scope at once.
Keys That Command Other People's Websites
The UpdraftCentral episode, also reported by BleepingComputer, makes the mechanism explicit. A threat actor breached TeamUpdraft's website, potentially compromising UpdraftCentral connection keys that can send commands to connected websites. Those keys exist to let site owners manage many WordPress installations from one place. In an attacker's hands, the same convenience becomes a remote command channel into every site that trusts the key.
This is the vendor-as-attack-surface pattern in its purest form. The compromised party is not necessarily the intended victim; it is the holder of credentials that others have already authorised. WordPress powers a large share of the web, and a substantial portion of US small businesses, publishers, and e-commerce operators run on it. A key that can push commands to connected sites is, functionally, a skeleton key held by a third party. When that third party is breached, customers inherit the risk without any change to their own code, passwords, or patching posture.
Agentic AI Lowers the Cost of the Middle Steps
The Wired reporting adds a third dimension: hackers around the world are all-in on using agentic AI to wage attacks, including manipulating tools like Anthropic's Claude to help carry out their crimes. AI guardrails, the story indicates, are not stopping criminal hackers.
The relevance to the vendor thread is indirect but important. Agentic tools excel at the repetitive, adaptive work that sits between initial access and impact: reconnaissance, credential handling, chaining steps together, and adjusting when something fails. That is exactly the labour that makes campaigns like SafePay's repeatable and attacks on shared keys scalable. If the marginal cost of the middle steps falls, the economics of attacking third-party infrastructure improve for criminals, and the number of organizations exposed through any single vendor breach rises.
