Cognition's Devin AI Swarm is now cleaning up the mess it created
Since Cognition and its AI friends have made writing code as easy as breathing, they have also turned the security landscape into a dumpster fire. Now, they are launching Devin Security Swarm to fix the very vulnerabilities their own agents keep spawning.
The logic is simple: if you generate 42% of the world's code using machines, you are bound to create a few security holes along the way. Companies are seeing a 10 to 100-fold increase in security alerts, mostly because AI is hallucinating code faster than humans can audit it. Cognition is now deploying its Devin Security Swarm to automatically scan for, confirm, and patch vulnerabilities before the bad guys find them.
Using an architecture they call Agentic MapReduce, the system runs parallel agents that dissect individual code segments while keeping an eye on the big picture. It effectively mimics an attacker by reproducing the exploits in an isolated sandbox. If the system manages to break into its own code, it automatically generates a pull request with a working patch.
In a benchmark against competitors like Claude Security, Codex Security, and Cursor Security, the Devin swarm successfully identified 72% of vulnerabilities. It even caught three critical bugs—including a tricky PHP sandbox escape and Spring Kafka deserialization issues—that every other tool completely missed. The company claims the cost per vulnerability detected is 30% lower than its closest rival.
It is truly a marvelous cycle of economic efficiency: generate bugs with one AI, bill the client for the security team, and then upsell them a second AI to patch the mess. Perhaps the logical endgame is a self-sustaining loop where algorithms just endlessly battle each other in a sandbox, leaving humans to simply watch the subscription fees tick upward into the void.
Source: Cognition
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.