ChatGPT solves complex cybersecurity task in hours, SentinelOne says

SentinelOne says ChatGPT completed a complex cybersecurity task in about 8 hours, a job that would typically take weeks for a human expert. The result centers on Gabriel Bernadett-Shapiro, whom SentinelOne describes as a distinguished AI research scientist, and it lands in a field where faster reasoning can change how quickly security teams spot, test, and respond to threats.

The finding matters because the task was not a simple prompt or one-off script. SentinelOne framed it as a multi-part problem that required sustained reasoning and many steps, the kind of work security analysts often spread across days or weeks. Bernadett-Shapiro’s LinkedIn profile places him at SentinelOne in San Francisco, California, and the company has been explicit that it is studying how frontier models behave when they are pushed beyond short answers into longer operational workflows.

That work has also focused on the offensive side of AI. On Nov. 3, 2025, SentinelOne posted a LABScon25 replay featuring Alex Delamotte and Bernadett-Shapiro, who described adversaries embedding large language models directly into malware to generate malicious code at runtime and evade traditional detection. SentinelOne has also published material on agentic AI, malware analysis, ChatGPT Security Risks: All You Need to Know, and ChatGPT & Cybersecurity - A Comprehensive Guide 101, underscoring that the company is treating AI as both a defensive tool and an attack surface.

The broader commercial setting adds pressure to that research. Reuters reported on Aug. 28, 2025, that SentinelOne raised its annual revenue forecast on strong cybersecurity demand, a reminder that the market for security software is growing even as the threats evolve. For defenders, the practical question is less about whether AI can help and more about where human oversight still matters, especially when a model can sustain complex cyber reasoning long enough to compress specialist work from weeks into hours.

That shift could favor analysts who use AI to accelerate triage, malware study, and threat hunting, but it also gives attackers more capability to automate the hard parts of malicious code generation and iteration. SentinelOne’s work points to a security environment in which speed is no longer the only metric; reliability, review, and judgment become the bottlenecks that decide whether faster AI strengthens defense or simply shortens the path to abuse.