AI found 10,000 serious bugs in a month, and fixing them didn’t scale. Here is what we think that means for the fixes stuck in your queue.
What Gartner published
In September 2026, Gartner® published two reports that name Reclaim Security as a Sample Vendor:
- Emerging Tech Impact Radar: Preemptive Cybersecurity (11 September 2026) names Reclaim a Sample Vendor for Preemptive Exposure Management.
- Agentic Security: Preemptive Exposure Management Demands AI-Driven Validation and Autonomous Interdiction (28 September 2026) names Reclaim among its Sample Vendors.
We also owe thanks to our customers. They run Reclaim in production, and their feedback shaped what we built. A Sample Vendor listing is not an endorsement, and we do not treat it as one. What interests us more is the direction the two titles share.
The constraint moved from finding to fixing
Both titles turn on the same word: preemptive. For more than a decade, our industry got very good at detection, then at prioritization. That progress means a security team now knows about more exposure than it can act on. In our view, the next era is preemption: neutralizing an exposure before an attacker uses it. That work starts after the finding, and the public data shows it is where programs fall behind.
- Discovery accelerated. In May 2026, Anthropic reported that, one month into Project Glasswing, it and about 50 partners had used Claude Mythos Preview to find more than ten thousand high- or critical-severity vulnerabilities. It also reported that “75 of the 530 high- or critical-severity bugs we’ve reported have now been patched.” Its summary: “Progress on software security used to be limited by how quickly we could find new vulnerabilities. Now it’s limited by how quickly we can verify, disclose, and patch the large numbers of vulnerabilities found by AI.”
- Remediation slowed. The 2026 Verizon Data Breach Investigations Report found that organizations fully remediated only 26% of the vulnerabilities in CISA’s Known Exploited Vulnerabilities (KEV) catalog in 2025, down from 38% the year before. The median time to full resolution rose from 32 days to 43. Vulnerability exploitation now starts 31% of breaches, the most common way in.
- Attackers stopped waiting. Mandiant’s M-Trends 2026 estimates the mean time to exploit in 2025 at minus 7 days, “meaning exploitation is routinely occurring before a patch is even released.”
Found fast. Fixed slowly.
Each dot is one high- or critical-severity bug Anthropic reported to maintainers through Project Glasswing. Filled dots had been patched by May 2026.
Found in the first month
More than 10,000
Reported to maintainers
530
Patched
75, about 1 in 7
Source: Anthropic, “Project Glasswing: an initial update,” 22 May 2026.
© 2026 Reclaim Security · reclaim.security
Finding is no longer the constraint. The fix is.
The last mile is a trust problem
Mobilization is the step between a prioritized finding and a deployed fix. It is also where exposures that everyone already knows about sit open for weeks.
Ask why a known-exploited flaw stays open for 43 days. In our experience, the answer is rarely laziness or bad prioritization; it is that nobody can say what the fix will break. The pattern is not new. In a 2021 Ivanti survey of more than 500 enterprise IT and security professionals, 61% said business owners ask for exceptions or push back maintenance windows once a quarter because their systems cannot be brought down. IT makes a rational call: when the cost of a change is unknown and the cost of waiting feels abstract, waiting wins. We made this case in more detail in Vulnerability remediation SLA: why risk-based tiers fail, and what to fix instead.
Closing that gap takes more than speed. An agent that can push a change at machine speed helps only when the people who own production trust the change. Trust, not speed, sets the pace of remediation.
Trust comes from concrete answers, delivered before the change moves: what it touches, what could break, who approves it, and how to roll it back. With those answers in hand, a team can extend automation one change class at a time, as its own record justifies it.
Four answers before the change moves
An example change request. It waits until each question has a concrete answer.
Example change request
Turn off legacy email sign-in
Approved. Rolling out to the first ring.
What does it touch?
Mail sign-in for 3 legacy apps and 140 shared mailboxes
UnknownWhat could break?
One scanner-to-email workflow. Its owner is notified.
UnknownWho approves it?
IT change board, at Thursday’s review
UnknownHow do we roll it back?
Re-enable in one step. The previous setting is saved.
Unknown4 of 4 answered.
© 2026 Reclaim Security · reclaim.security
How Reclaim works the last mile
Reclaim Security is a Unified Exposure Management Platform. It drives mobilization across the stack a customer already owns, and it owns the entire remediation life cycle, from assessment to deployed change. Reclaim neutralizes exposures by changing the configurations and policies of the controls a customer already runs, which complements patching rather than replacing it. Each stage answers a question that usually stalls a fix:
Assessment: does this exposure matter here?
Reclaim judges each exposure by susceptibility, not severity: against how the organization actually works and which of its controls already stand in the way.
Planning: which change neutralizes the most?
SC™ (Susceptible Components) maps exposures to the components they share, so one remediation can close a whole batch.
Execution: what will this change break?
PIPE™, the Productivity Impact Prediction Engine, plans valid policies and predicts the business impact of a change before anything moves. Rollout runs in rings, stays reversible, and proceeds fully or semi-automatically at the customer’s pace.
© 2026 Reclaim Security · reclaim.security
Reclaim's AI Security Engineer monitors and controls all three stages. The customer's team decides which changes run on their own and which wait for a person to approve them. The change that runs against production is deterministic and policy-bound: the change a team approves is the change that runs.
The record so far: across more than 60,000 automated remediations in the trailing 12 months (as of May 2026), Reclaim recorded no production-impacting changes. That is a measured history, not a promise, and it is the kind of record the last mile runs on.
The second report's title also names AI-driven validation and autonomous interdiction. Reclaim's work sits in the step that turns a decision into a deployed change: making the fix itself safe enough to ship.
Where exposure management goes next
Here is our view of the next few years, and what it means for security leaders:
- The closed exposure becomes the unit of progress. Programs will report exposure debt retired, net of new findings, rather than findings discovered.
- Change safety becomes a metric. The share of changes a team ships without disruption predicts how fast it can remediate everything else.
- Autonomy arrives in increments. Teams will extend automation change class by change class, as their own record earns it, not through a single switch.
One step worth taking now: count the fixes that stalled last quarter because of change risk rather than priority. That number shows how much of your backlog is a trust problem, and it gives security and IT a shared place to start.
How much of your backlog is a trust problem?
Count the fixes that stalled last quarter, then the ones that stalled because of change risk rather than priority. Divide the second number by the first.
Stays in your browser.
Example numbers
60% of your stalled fixes are a trust problem
Change risk12
Priority or other8
© 2026 Reclaim Security · reclaim.security
Finding exposures stopped being the hard part. The hard part is the moment someone has to approve a change without knowing what it will touch. We built Reclaim for that moment, and we're grateful that Gartner named us in two reports on preemptive security.
Barak Klinghofer, CEO, Reclaim Security
Chat with this post in your AI assistant
A source-grounded recap of this post on preemptive exposure management and why fixes stall at the last mile. Ask how it applies to your stack, anchored to what the post actually says.
Works with ChatGPT, Claude, Gemini, or any LLM. It stays grounded to this post, with its sources.
© 2026 Reclaim Security · reclaim.security
Sources
- Anthropic, "Project Glasswing: an initial update" (22 May 2026)
- Verizon, 2026 Data Breach Investigations Report, Executive Summary (May 2026)
- Mandiant, M-Trends 2026 (23 March 2026)
- Ivanti, patch management study via Business Wire (6 October 2021)
- Reclaim Security, Vulnerability remediation SLA: why risk-based tiers fail, and what to fix instead
Gartner, Emerging Tech Impact Radar: Preemptive Cybersecurity, Elizabeth Kim, Charanpal Bhogal, et al., 11 September 2026.
Gartner, Agentic Security: Preemptive Exposure Management Demands AI-Driven Validation and Autonomous Interdiction, Luis Castillo, Elizabeth Kim, et al., 28 September 2026.
GARTNER is a trademark of Gartner, Inc. and/or its affiliates.
Gartner does not endorse any company, vendor, product or service depicted in its publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner publications consist of the opinions of Gartner's business and technology insights organization and should not be construed as statements of fact. Gartner disclaims all warranties, expressed or implied, with respect to this publication, including any warranties of merchantability or fitness for a particular purpose.



