Torq is adding a new layer of intelligence to its proprietary AI SOC Platform. SOC Brain learns from past incidents, investigations, and daily operations to better adapt to the organization. The result is intended to deliver a solution that is more than just a simple LLM wrapper.
Torq is aware that many competitors have already claimed this “self-learning” capability for their own AI applications. The difference with SOC Brain is that this solution relies on more complex information. According to Torq CEO and co-founder Ofer Smadari, alternative security tools essentially just forward cases to an LLM for isolated decisions. Torq, on the other hand, reportedly draws on information that predates its installation by a significant margin.
Recall, Reflex, Retrospect
Three components work together to achieve this security goal of a self-learning AI. First, there’s Torq Recall, which analyzes historical data, from logs and traces to analyst notes, conflicting decisions, and the significance of certain “observables” such as IP addresses, URLs, and hostnames. What was once “tribal knowledge,” Torq argues, is transformed into institutional knowledge.
Torq Reflex goes a step further and trains AI models based on the SOC’s behavior and output. It serves as a kind of assessment of the risks a particular SOC has accepted, where it took decisive action, and how it performed triage. In 85 percent of cases, Torq Reflex’s decision is said to be identical to that of human SOC analysts.
Finally, there is Torq Retrospect, the truly “learning” aspect of these three components. Before it is deployed, SOC Brain can learn from historical examples and forensically analyze resolved incidents. This component is what makes information from other security solutions available to Recall and Reflex.
Isolation
Torq emphasizes that this information remains within the organization itself. Although SOC Brain learns everything from the daily practices and challenges of security teams, it is confined to this group. The default configuration is therefore isolated, with options for data governance where necessary. Support for the EU AI Act is also built in.