The number of software vulnerabilities discovered in 2026 is on pace to nearly double the total for 2025, fueled by artificial intelligence tools that find security flaws at machine speed. Data from the US National Vulnerabilities Database shows 45,207 security flaws recorded between January and July 27 — already approaching the full-year record set in 2025.
Oracle patched 1,449 vulnerabilities in its July Critical Patch Update, the largest security release in the company’s 49-year history. Microsoft disclosed 642 bugs in its July Patch Tuesday, another all-time high and nearly five times the July 2025 count. Google found and fixed 433 flaws in a single Chrome update, compared to just 11 in the equivalent update last year, as reported by Bloomberg.
“We have to come to the reckoning that these tools are increasing the ability of people to find vulnerabilities in software,” said Gabriel Bernadett-Shapiro, distinguished AI research scientist at SentinelOne, in the Bloomberg report.
AI Models Are Reshaping Vulnerability Discovery
The explosion in discovered vulnerabilities stems directly from advances in frontier AI models. Anthropic’s Mythos tool found thousands of software vulnerabilities in early testing, showcasing a new level of capability for modern AI. Google’s Chrome engineering director Doug Turner attributed the “unprecedented scale and speed” of vulnerability discovery to advances in AI models and corresponding investment by the company.
Microsoft released a new AI security tool called MAI-Cyber-1-Flash on Monday, designed to help organizations manage the growing flood of software vulnerabilities. The tool arrives as the average time attackers take to turn an abstract vulnerability into a working exploit has dropped from 72 hours in 2025 to just 24 hours in 2026, according to Alexander Leslie, senior adviser at Recorded Future.
OpenAI disclosed on July 21 that its autonomous agents breached another company — Hugging Face — in an incident that took hours compared to the weeks it would likely have taken a human operator. The same week, Anthropic published research showing its Claude Mythos Preview model discovered cryptographic weaknesses in HAWK, a post-quantum signature candidate, and a reduced-round version of AES-128 after approximately 60 hours of autonomous work, as documented in The Hacker News.
Yet the picture is not as dire as the raw numbers suggest. According to the US government’s Known Exploited Vulnerabilities catalog, there has been no corresponding rise in the number of flaws actually exploited in the wild so far in 2026, despite the dramatic uptick in discovered vulnerabilities. Most of the new findings come from internal security teams using their own AI-powered tools, not from malicious hackers.
“We just aren’t seeing the numbers to back up the doom and gloom prophets,” said Dustin Childs, head of threat awareness at Trend Micro, as quoted by Bloomberg.
Fastjson 1.x RCE: An Active Threat With No Patch Available
While the overall exploitation rate has not spiked, specific vulnerabilities demand urgent attention. One of the most critical active threats this week is CVE-2026-16723, a remote code execution vulnerability in Alibaba’s Fastjson 1.x library affecting versions 1.2.68 through 1.2.83. The flaw carries a CVSS score of 9.0 and allows unauthenticated attackers to execute arbitrary code with the privileges of the Java process.
The exploit chain works through a Spring Boot executable fat-JAR and does not require AutoType enablement or a classpath gadget — making it much easier to weaponize than typical Java deserialization vulnerabilities. Security firms ThreatBook and Imperva have observed active exploitation targeting the financial, healthcare, and retail sectors in the United States, Singapore, and Canada.
Because no official patch has been released for the Fastjson 1.x branch, organizations must use mitigation strategies: enabling SafeMode via the JVM parameter -Dfastjson.parser.safeMode=true, switching to the restricted 1.2.83_noneautotype build, or migrating to the unaffected Fastjson2 library. The F5 Labs weekly threat bulletin for July 29 covers this in detail at F5 Labs. Security teams should also audit their software composition for any applications using vulnerable Fastjson versions.
Hotel Wi-Fi DNS Hijacking Campaign
In a separate active campaign, threat actors are compromising Wi-Fi gateways at hotels and conference centers globally to hijack DNS settings and redirect users to fraudulent Microsoft 365 login pages. The attackers modify DNS configurations to point to malicious domains such as m365-owa[.]com and owa-ms365[.]com, then trick victims into authorizing a device-code authentication flow that bypasses multi-factor authentication.
The campaign has been active since at least June 2026 and targets traveling corporate employees across financial, legal, healthcare, and energy sectors. Organizations can mitigate this threat by enforcing always-on full-tunnel VPNs, using encrypted DNS in strict mode, disabling Web Proxy Auto-Discovery, and disabling the Device Code authentication flow in Microsoft Entra ID. BleepingComputer reported the full technical breakdown at BleepingComputer.
Record Patch Cycles Test Enterprise Defenses
The record-breaking patch cycles from Oracle, Microsoft, and Google signal a structural shift in how software vulnerabilities are found and fixed. Oracle’s July 2026 Critical Patch Update alone addressed 1,449 flaws across 334 products, including 600 remotely exploitable vulnerabilities. Many of those flaws were likely discovered with the help of AI tools. The Cyber Express covered the scale of the update at The Cyber Express.
For security operations teams, the implications are straightforward: the volume of patches will continue to grow as AI tools improve. Organizations need automated patch management pipelines, software composition analysis integrated into CI/CD workflows, and risk-based prioritization to separate the critical vulnerabilities from the noise.
The semiconductor and cloud computing sectors are particularly affected, given their reliance on complex software supply chains. A vulnerability in a widely used library like Fastjson can cascade across hundreds of products, as cybersecurity researchers at F5 Labs noted in their recent threat bulletin.
Additional Active Threats This Week
Beyond the Fastjson vulnerability and hotel Wi-Fi campaign, several other threats are active. The Chaos ransomware group’s msaRAT hides its command-and-control traffic inside legitimate headless Chrome and Edge browser processes. The malware uses the Chrome DevTools Protocol to inject JavaScript and route double-encrypted traffic through Twilio TURN relays and Cloudflare Workers, making the malicious traffic indistinguishable from normal web activity.
An unpatched zero-day vulnerability in the Cursor AI coding assistant allows attackers to hijack Windows developer machines through a malicious git.exe binary placed inside a cloned repository. Cursor prioritizes the root of the workspace when searching for the Git binary, so dropping a renamed executable into a repository triggers silent automatic execution. Developers should inspect cloned repositories for unexpected executables before opening untrusted projects.
Redis released emergency security updates to patch two critical memory corruption vulnerabilities that Kimi K3 AI agents reportedly identified and weaponized into functional exploits in under 90 minutes. The vulnerabilities enable authenticated remote code execution through Redis Streams and the TDigest RDB loader in the RedisBloom module. While no in-the-wild exploitation has been detected, administrators should upgrade immediately or revoke RESTORE privileges.
The Defender’s Challenge
The core challenge, according to SentinelOne’s Bernadett-Shapiro, is that AI has democratized vulnerability discovery. Both defenders and attackers now have access to tools that can find flaws faster than human researchers ever could. The difference lies in who can patch faster and deploy countermeasures more efficiently.
Not all the news is alarming. Trend Micro’s Childs pointed out that the Known Exploited Vulnerabilities catalog shows no surge in actual exploitation, suggesting that the security industry’s defensive AI tools are keeping pace. Microsoft’s new MAI-Cyber-1-Flash model and Google’s internal AI-powered vulnerability hunting programs are examples of defenders using the same technology to stay ahead.
The write speed matters. In 2025, attackers needed an average of 72 hours to weaponize a vulnerability. In 2026, that window has dropped to 24 hours, according to Recorded Future. Security teams that rely on monthly patch cycles are falling behind. The organizations best positioned to handle the new reality are those investing in automated patch management, AI-augmented defense tools, and real-time threat intelligence feeds.
As the second half of 2026 unfolds, the trend is clear: AI will continue to accelerate vulnerability discovery, and both sides of the cybersecurity equation will need to adapt. The ones that rely on manual processes and periodic patch cycles will struggle to keep up.