Cisco Partner Summit: Exploit Window Shrinks to Eight Hours as AI Agents Start to Act

Exploit windows have collapsed

According to Kozyrkov's recap, Tim Coogan, Cisco's SVP of Global Partner Sales, laid out how quickly the security timeline has compressed. He said: "In 2018, there were more than two years between when a vulnerability was discovered and when it was exploited. Last year, it was about three weeks. This year, it's eight hours or less."

Kozyrkov drew a practical conclusion from the figure. With only eight hours, she wrote, there is no time to work out a response after an incident begins, so leaders need to agree on their default actions ahead of time. Coogan tied the point to AI adoption: "Customers cannot adopt AI until they trust the systems that run it."

From answers to actions

Chuck Robbins, Cisco's Chair and Chief Executive Officer, spoke about AI agents working around the clock. In the recap he is quoted as saying: "these agents don't stop and go home and have dinner. They work 24/7, and there's no limit to how many you can have working." He also named the concern customers are weighing: "Trust... takes years to earn and moments to lose."

Srini Namineni, Cisco's Chief Automation Officer, asked the audience to raise a hand if they had used AI in the last 24 hours. Nearly everyone did. He then asked them to keep a hand up only if AI had actually done something for them, such as sending an email or booking a meeting, and about half the hands went down. His takeaway, as quoted: "We're used to getting answers from AI, but we're moving to an era where AI can actually do things for you."

Kozyrkov argued that getting an answer and letting AI act are very different decisions. Once AI acts, she wrote, someone must decide what it is allowed to do and who is accountable when something goes wrong.

Cisco's own numbers on support and security

Namineni also shared a result from Cisco's own operations. The company cut its partner support cases from 4 million a year to 400,000. "By fixing the root causes, we have reduced them by 90 percent," he said. Kozyrkov's reading is that organizations should fix the reasons people open cases first and then use AI to close the remainder faster. She said she would recommend that order to anyone planning to automate a process.

Liz Centoni, Cisco's EVP and Chief Customer Experience Officer, opened her talk with "Four days to 10 minutes." According to the recap, a global insurer used to need four days to find out which of its 16,000 devices were affected by a single security alert. It now needs 10 minutes.

What it means for customers and partners

The recap is a personal summary rather than an official Cisco release, and Kozyrkov marked the post as sponsored by Cisco. The figures are the speakers' own claims from the keynote, and the post does not provide further detail on how they were measured.

Even so, the themes are clear. Security teams are being told to plan responses before an incident, because the time to react is shrinking. Companies considering AI agents are being pushed to define in advance what those agents may do and who answers for the outcome. Commenters on the post raised similar points, calling for pre-agreed boundaries, clear accountability and human checkpoints before agents change records or close cases. For Cisco's partners, the message from the stage was that trust is a precondition for adopting AI, and that fixing underlying problems comes before automating around them.