AI cybersecurity threats: Microsoft Digital Defense Report highlights rising risks for defenders

Faster exploits and a growing AI attack surface leave cybersecurity teams racing to keep up.

AI cybersecurity threats are creating new challenges for defenders, from faster vulnerability exploitation to attacks targeting AI workloads and agents. Microsoft’s Digital Defense report provides insights into how cybersecurity threats are evolving in the age of artificial intelligence (AI), including the pressures facing teams responsible for vulnerability remediation, phishing defense and ransomware protection.

Key takeaways

• Vulnerability exploitation is outpacing remediation, leaving defenders little time to respond.
• AI workloads and agents expand the attack surface, adding risks involving malicious links, data exfiltration and compromised credentials.
• Phishing, impersonation, and ransomware continue to add pressure to cybersecurity teams.

Based on an analysis of more than 165 trillion security signals that Microsoft tracks daily, the report notes that the median discovery-to-weaponization time is now well below 24 hours. The challenge is median enterprise remediation for critical external Common Vulnerabilities and Exposures (CVEs) ranges from 30 to 60 days while median time-to-exploit is only about five days. In fact, the report notes median time from container start to first exploit attempt is 5.3 hours, and more than half of compromised containers are exploited within 24 hours.

It’s not clear to what degree AI is already helping cybercriminals shorten the time needed to weaponize vulnerabilities. The general expectation, however, is that reverse engineering a vulnerability to create an exploit will soon take only a few minutes. If that expectation holds, median time to exploit could fall further as more attacks are launched at machine speed using advanced AI models that are becoming more accessible with each passing day.

AI workloads and agents expand the attack surface

The report also notes that more cyberattacks are already being aimed at various classes of AI workloads, with malicious link injection/exfiltration representing 52% of all observed AI-workload attacks. Those types of attacks are only likely to increase in the agentic AI era with 82% of organizations planning broader rollout within 12 to 18 months. In total, about 1.3 billion agents are projected in production by 2028.

Thus far, attacks on AI agents have primarily involved malicious link injections or attempts to violate policies (45%), compared to sensitive-data exfiltration (17%) and system-prompt leakage (14%). As more AI agents are deployed, however, cybersecurity teams should assume adversaries will first attempt to steal the credentials used to invoke them before then ultimately trying to insert a malicious agent they created into a workflow. Regardless of the tactics and techniques used, the one thing that is certain is the overall level of risk is rising because AI agents once compromised are able to perform any number of tasks that could result in a major security breach.

The issue, of course, is AI agents and workloads are expanding an attack surface that most cybersecurity teams are already struggling to defend. The Microsoft report noted that an estimated 72,000 additional CVEs will be discovered in 2026. While not every CVE may be especially relevant to every IT environment, the overall cognitive load stemming from what are now hundreds of thousands of CVEs is substantial.

Phishing, impersonation, and ransomware add pressure

Just as challenging, the report notes CAPTCHA-gated phishing exceeded 100 million attacks in the last year, while impersonation signals appeared in nearly 85% of reported phishing emails. Business contact impersonation exceeded 46 million attacks in the last 12 months, while confirmed malicious Teams call volume rose 502% year over year.

Finally, Microsoft Defender recorded a 15.8% year-over-year increase in enterprise ransomware detonations.

Will AI tools and platforms ultimately help cybersecurity teams more than attackers? That remains the hope, but defenders are still waiting for the balance to shift. In the meantime, cybersecurity professionals face relentless pressure, and time is not on the side of the defender.