The record-breaking Patch Tuesday release—which may cover more than 600 CVEs (Common Vulnerabilities and Exposures) in total—shows that the era of ‘AI against human is done,’ one MSP executive tells CRN.
Microsoft’s unprecedented July security update—which potentially includes patches for more than 600 vulnerabilities—is confirmation that ultra-powerful AI models for discovering software flaws are already starting to signal massively bigger challenges in vulnerability management, according to MSP executives.
For months, the industry has been watching for indicators that vulnerability discovery capabilities from frontier AI models such as Anthropic’s Claude Mythos would lead to increased CVEs (Common Vulnerabilities and Exposures). While some hints of this had surfaced previously, the hundreds of CVEs that received software patches Tuesday as part of Microsoft’s monthly release of software bug fixes is the clearest sign of this yet, MSP executives told CRN.
[Related: Why Frontier AI Models Are Creating A Vulnerability ‘Storm’: Apiiro CEO]
Estimates of the number of vulnerabilities disclosed by Microsoft during the release, colloquially known as “Patch Tuesday,” have varied but could be above 600, according to TrendAI’s Zero Day Initiative. Dustin Childs, head of threat awareness at TrendAI, counted 621 new Microsoft CVEs for the July security update.
“The bug apocalypse has fully descended upon us,” Childs wrote in a post.
By comparison, Microsoft’s June release had covered 206 CVEs, which had itself been the record at the time. Prior Patch Tuesday releases from Microsoft, including as recently as earlier this year, often did not exceed 100 CVEs for the month.
The number of Microsoft products receiving patches in the July release is also “astonishing,” Childs wrote, and it’s notable that 63 of the CVEs are rated “critical” in terms of severity, as well. Just two of the vulnerabilities are listed as under active exploitation, however, he noted.
Microsoft has been among the tech industry vendors that have received access to Anthropic’s Claude Mythos model for the purposes of proactively discovering vulnerabilities in its own products, as part of the Project Glasswing initiative announced in April. Similarly, OpenAI has provided Microsoft with access to its frontier AI vulnerability-discovery models through its Trusted Access for Cyber initiative.
Speaking with CRN Tuesday, MSP executives said that there should be little doubt that AI is the biggest factor that has changed in discovery of vulnerabilities in Microsoft products.
Without question, “this is being driven by the more powerful models being readily available,” said Travis Woods, strategic executive advisor at Fort Point IT, a Novato, Calif.-based MSP. “It’s obviously accelerating the defensive stance that folks are having to take. And it’s AI against AI. Because AI against human is done.”
For IT and security teams, the huge number of vulnerabilities will make it more essential than ever to prioritize effectively around which patches to deploy first, executives said.
“It really changes the way that we approach patching and security,” Woods said. “The priority of the CVE is not enough information to make a good determination of what gets patched first.”
Ultimately, “we’re going to have to pull in more ancillary data to help prioritize what’s first,” he said.
While the first priority should always be to address vulnerabilities that are being actively exploited, MSPs will then need to evaluate which of the remaining flaws pose meaningful risks within a given customer’s environment, according to Zack Finstad, vice president of cybersecurity at Logically, No. 332 on CRN’s Solution Provider 500 for 2026.
And organizations will also have to become more comfortable with the fact that not all vulnerabilities will be able to be addressed right away, Finstad said.
The key for many organizations and MSPs will be “how to evaluate the risk associated with not dealing with [CVEs] right away,” he said.
The massive scale of the release also suggests that organizations could face difficult decisions about whether to shorten the testing traditionally performed before patches are deployed to customer systems, Finstad said.
“We may not have the time to do the diligence that we’ve always done—testing and validating on our own systems, in our labs, before rolling out to customers,” he said. “We may just have to buckle down, patch and deal with the fallout that comes out afterwards.”
At the same time, the disproportionately modest number of known exploited vulnerabilities disclosed Tuesday means the industry is not seeing an immediate surge in attacks from the increased bug discovery activity at this point, Finstad noted.
However, many experts believe that it’s only a matter of time before threat actors will have access to similarly potent AI capabilities for vulnerability discovery and exploitation—if they don’t already.
A key question will increasingly be, “how do we protect ourselves against [threat actors’] ability to use tools that we also have access to?” Finstad said.
All in all, the developments mean that patching will less and less be about processing a monthly list of vulnerabilities, and more about continuously determining which risks matter most for each customer, Woods said.
“We can still only patch so much,” he said.
