Good morning, I'm your AI Brief anchor. Here's what's happening in AI today, Tuesday, April 28, 2026.
Cognition Labs Hit by Major 'Model Inversion' Attack
Our top story this morning: a major security breach at Cognition Labs, the high-profile startup behind the AI software engineer, Devin. In a statement released overnight, the company confirmed that a sophisticated threat actor successfully stole significant portions of Devin's proprietary codebase.
This wasn't a typical network intrusion. Security analysts say the attackers used a novel technique known as a "model inversion" attack. Instead of breaching servers to steal files, this method essentially reverse-engineers the AI model itself. By carefully analyzing the model's outputs to specific inputs, the attackers were able to reconstruct parts of the underlying source code and architecture that make Devin so powerful.
The incident sends a shockwave through the industry, raising critical new questions about how to protect the intellectual property at the core of these multi-billion-dollar AI systems. When a company's "secret sauce" is an AI model, protecting it from digital espionage becomes an entirely new challenge. Cognition Labs says it is working with federal authorities and has already deployed countermeasures, but the damage may already be done.
White House Issues Landmark AI Security Mandate
The breach at Cognition Labs underscores a growing concern that has now prompted action from the highest levels of government. In a landmark move, the White House has issued a new Executive Order aimed at securing the nation's most critical AI systems.
Titled the "AI System Resilience and Security Act," the order establishes the first mandatory cybersecurity standards for AI deployed in crucial sectors. This includes finance, energy, healthcare, and transportation. Under the new rules, companies operating these systems will be required to conduct rigorous "red teaming"—a process of simulated attacks to find vulnerabilities—before deployment. They will also be mandated to report any discovered vulnerabilities to a newly formed government oversight body.
Administration officials stated that the goal is to create a baseline of security and accountability for AI that has the potential to impact national security and public safety. The order signals a major shift from industry self-regulation to direct government oversight, a move that is being closely watched by AI developers and civil liberties groups alike.
Stanford Researchers Unveil 'Cognitive Jigsaw' Jailbreak
As the government works on policy, new research reveals just how fast the threat landscape is evolving. Researchers at the Stanford AI Lab have published a paper detailing a powerful new jailbreak technique they call "Cognitive Jigsaw," and it appears to bypass the safety alignments in all major large language models.