The Exploit Window Is Now Negative: Mandiant M-Trends 2026 Confirms Patch-Based Defense Is Dead

TL;DR

Mandiant's M-Trends 2026 report—based on 500,000+ hours of 2025 incident investigations—reveals that the mean time to exploit a vulnerability has dropped to negative 7 days. Attackers are weaponizing exploits before patches are released. Traditional vulnerability management is structurally obsolete.

What Happened

Google Cloud's Mandiant team released the M-Trends 2026 report in early May 2026, delivering findings from frontline incident response investigations across hundreds of breaches conducted throughout 2025. The headline is brutal: the assumption that defenders can patch before attackers strike has completely broken.

The mean time to exploit vulnerability has collapsed to -7 days—meaning that exploits are actively deployed in the wild for an average of seven days before patches exist. This is not a marginal regression. This is a structural inversion of the vulnerability lifecycle that has governed incident response doctrine for 30 years.

The backup statistics are equally grim:

• Median dwell time for cyber espionage groups: 122 days
• Time from initial breach to ransomware deployment: collapsed from 8+ hours (2022) to 22 seconds (2025)
• Exploits as initial infection vector: 32% (sixth consecutive year of dominance)
• Malicious packages in public repositories: 454,600 in 2025 (up 726% from 55,000 in 2022)
• Vulnerabilities in large orgs never remediated: 45%
• Average remediation time for critical/high severity: 74 days

Read against a mean time to exploit of -7 days: the gap is not a patch cycle. It's structural collapse.

Technical Details

The AI Accelerant

Mandiant's analysis isolates AI as the primary time-compression agent. This is not AI causing breaches directly—the underlying failures remain human and systemic. But AI has collapsed every stage of the attack lifecycle:

• Reconnaissance: LLM-assisted target identification and social engineering at machine speed
• Phishing & social engineering: AI-generated, contextually convincing content that human analysts struggle to distinguish from legitimate communication
• Malware development: Functional, stealthy malicious code generated from natural language prompts
• Evasion & adaptation: Malware families actively querying LLMs during runtime to evade detection; credential stealers scanning compromised machines for local AI tools
• Supply chain poisoning: Malicious npm/PyPI packages grown from 55,000 (2022) to 454,600 (2025)—the sharpest jumps aligned with GPT-4 release (2023) and agentic coding tools (2025)

One finding stands out: AI-generated malicious code now passes static analysis and signature scanners that organizations have relied on for years.

The Democratization of Sophisticated Attack Capability

Mandiant documented three specific incidents that crystallize the shift:

1. February 2025: Three teenagers with no coding background used an LLM to build a tool that targeted Rakuten Mobile over 220,000 times.

2. July 2025: A single actor using agentic AI tools conducted extortion against 17 organizations in one month—automating code development, data analysis, and ransom communications.

3. December 2025: One individual used AI coding tools to breach 10+ Mexican government agencies and exfiltrate 195 million taxpayer records.

The technical barrier to conducting sophisticated attacks has collapsed. The population of capable threat actors is expanding rapidly because the labor cost of exploitation has fallen to near zero.

Lyrie Assessment

This report lands at the exact moment traditional defenders have been admitting they've lost. Mandiant's -7-day metric is not a prediction—it's a measurement of what is already happening.

For Lyrie's audience (CISOs, security engineers, infrastructure defenders), the implications are severe and concrete:

1. Patch management as a primary control is dead. Waiting for patches is playing defense in an era where exploit development is now a 48-72 hour turnaround, not a 6-month R&D cycle. Organizations remediating critical vulnerabilities in 74 days are remediating in the past tense—the breach has already occurred.

2. The attack surface itself must shrink. You cannot outrun attacks where attackers have a 7-day head start. Mandiant's recommendation is explicit: stop trying to patch faster. Instead, eliminate entire categories of vulnerability by reducing the attack surface. Virtualization, unmonitored Tier-0 assets, and SaaS integration points are now breach doorways, not business enablers.

3. Autonomous response is no longer optional. A 122-day dwell time for espionage and a 22-second time-to-ransomware window means human-driven incident response is measuring in the wrong unit. If initial compromise to adversary objectives takes 22 seconds, manual detection and response cannot participate in that timeline. Autonomous defense systems that can detect and respond in milliseconds are now a hard requirement, not a "nice-to-have."

4. The skills gap closed on the wrong side. When teenagers and solo operators can conduct nation-state-grade attacks using LLMs, the adversary population has expanded exponentially. This is not a temporary phase. This is the new normal.

5. Supply chain poisoning at scale is here. 454,600 malicious packages means that dependency management—the daily act of pulling third-party code into your build pipeline—is now a primary attack vector. Every npm install, pip install, or cargo fetch is a potential privilege escalation.

For Lyrie's mission: this report validates that autonomous cyber operations are now a defensive necessity, not a curiosity. Organizations that do not deploy autonomous detection, response, and remediation at machine speed will continue to operate with 122-day dwell times and 22-second-to-objective timelines. That is not risk management. That is losing.

Recommended Actions

1. Immediately audit your attack surface. Not your patch backlog—your surface. What systems are exposed to the internet? Which SaaS integrations run unmonitored? What Tier-0 infrastructure is still unguarded? Shrink it.

2. Deploy automated response capabilities. Your SOC cannot respond in milliseconds. Automation can. Implement detection rules that trigger autonomous containment (network isolation, process kill, credential revocation) without human approval loops.

3. Tier your vulnerability response by exploitability, not severity. A CVSS 5.0 with active public exploits is more urgent than a CVSS 8.0 with no known attack path. Mandiant's data shows 32% of breaches start with known exploits. Hunt for them specifically.

4. Assume supply chain compromise is happening now. Implement dependency scanning at build time, not release time. Verify all third-party code signatures. Isolate build environments from the internet.

5. Measure dwell time obsessively. If your dwell time is still measured in weeks, you have not detected the breach—you have detected the aftermath. Invest in behavioral anomaly detection that catches lateral movement in hours, not days.

6. Build for operational continuity under compromise. Mandiant's data on espionage dwell times (122 days) means assume attackers are inside your infrastructure right now. Architect for zero-trust execution, immutable logs, and autonomous resilience.

Sources

1. Analytics Drift: AI Made Cyberattacks Faster Than Patches. Mandiant's Data Proves It.

2. Google Cloud Mandiant: M-Trends 2026 Report Landing Page

---

Lyrie.ai Cyber Research Division