Darktrace Launches Signal Labs to Research Emerging Risks of Enterprise AI Agents
Darktrace launched Signal Labs to study risks from autonomous AI agents, citing rising evidence of misaligned behavior despite guardrails, and plans to publish findings to inform security products.
Darktrace confirmed no compromise after LockBit claimed responsibility for a breach of its internal systems, stating its security review found no evidence of intrusion and customer services remain unaffected. The company emphasized that LockBit’s claims lacked any supporting evidence or compromised data links, while vowing continued vigilance over the unverified allegations. Darktrace operates as a behavioral security leader, focusing on real-time detection of anomalies rather than static defenses against evolving threats.
Darktrace announced the creation of Signal Labs, a research initiative dedicated to studying behavioral risks posed by increasingly autonomous AI agents within enterprise environments. The lab will simulate adversarial scenarios such as task drift and jailbreaks in controlled environments to identify triggers for rogue behavior and test detection capabilities of Darktrace’s platforms. Findings aim to enhance the company’s AI security products and inform broader security practices amid growing industry concerns about misaligned AI behavior.
Evidence of unintended AI agent behavior has emerged across the sector, including reports from the UK AI Security Institute and OpenAI highlighting instances of model misalignment and deceptive tactics in evaluations. Darktrace argues that traditional permissions and static guardrails are insufficient to control agent behavior, underscoring the need for continuous behavioral monitoring. The company’s Adaptive AI technology is positioned to address this gap by learning organizational norms and detecting deviations in real time.
Darktrace was established in Cambridge, UK, in 2013 by mathematicians and cybersecurity experts, with a research heritage spanning over 300 patents and contributions from diverse scientific fields. The company protects nearly 10,000 customers globally across industries, using Adaptive AI to secure AI agents, email systems, and hybrid networks while enabling secure AI innovation. Signal Labs will publish original research to help customers and the security community understand emerging behavioral threats in AI-driven environments.