Sophos resolves half its MDR cases end to end with AI. The customer's response mode sets the agents' authority
Sophos cut response time for agent-handled cases from about 38 minutes to 89 seconds with OpenAI Daybreak. The transferable part: authority set in the service contract, for people and agents alike.
Listen to this article · 5 min
AI-generated narration of the full article.
At 2 a.m., a detection fires on a customer’s server. Who is allowed to isolate that machine: the customer’s own team, a security provider’s analyst, or an agent working for that provider?
At Sophos, the answer was settled before any agent existed, and that is why its agents could be put to work quickly.
OpenAI published the case on October 9. Sophos runs Fusion, its AI-native defense system that includes Managed Detection and Response, across sensor data from more than 500 third-party integrations and its own products. Those sensors produce trillions of events a day, which Sophos distills into roughly 1,000 to 2,000 cases for its nine security operations centres.
The operating model before the agents
Investigating and responding to a case depended mostly on human expertise and averaged about 38 minutes, a figure CTO John Peterson says was better than 96% of professional SOCs. The constraint was not quality. It was scarce analysts against rising volume.
The architecture now
Agents built through OpenAI’s Daybreak program changed how cases move. An investigation agent gathers the customer context, detections, indicators of compromise and relevant threat intelligence for each case. A planning model runs a plan, execute and review loop: it builds an investigation plan, completes the steps and produces a summary with recommended response actions for analysts to review. Other agents can carry out parts of the response.
Sophos reports that average response time for cases using these agents fell to about 89 seconds, and that 52% of MDR cases are resolved end to end by AI, “within boundaries calibrated by Sophos analysts.” These are Sophos’s figures for its own workloads, published in a vendor’s customer story.
The boundary was already in the contract
Sophos’s MDR service has three operating modes, chosen by the customer:
Who acts on a case, by the customer's choice
Customer actsSophos acts
- NotifySophos investigates and recommends; the customer acts
- CollaborateSophos and the customer agree before action
- AuthoriseSophos responds directly on the customer's behalf
The key sentence in the case is this one: “The same boundaries apply whether work is completed by a person or an agent.” An agent working a case for a Notify customer can investigate and recommend; it cannot act, because a Sophos analyst could not act either. Potentially destructive actions still require the right level of human oversight, and, in Peterson’s words, “Anything we don’t feel comfortable with an agent handling gets passed off for human judgement.”
That is why 52% end to end is a meaningful number rather than an alarming one. The agents did not receive a new grant of authority. They inherited an existing one, defined in a service contract that customers had already accepted and that analysts already worked within.
What transfers to other operations teams
The pattern we saw in Exabeam’s agentic SOC and in Cornerstone’s database agents holds here at larger scale: agents reach production fastest where the boundary of action was defined before they arrived.
Most internal operations teams do not have that boundary written down. Who may restart a production service, revoke a user’s sessions or quarantine a mailbox is often a matter of habit and seniority. Before an agent can act on those things, the authority has to exist as a rule, scoped by system and by the business owner’s choice, the way Sophos scopes it by customer.
Three practical steps follow:
- Write the modes first. For each class of response, decide who may act and under what agreement, independent of whether a person or an agent does the work.
- Let the system owner choose the mode, as Sophos customers do, rather than letting the operations team set its own ceiling.
- Measure automation within the mode. “Half of cases resolved end to end” is only meaningful next to the boundary it was resolved within, which is also how autonomy is earned: per action class, on evidence.
Peterson’s closing advice to security leaders was not about AI at all: patch, use multifactor authentication, segment networks and run strong operations. The agents make Sophos faster. The boundaries it built long before them are what make the speed safe.