Security1 publisher2 min readPublished
Exploitation catalogues logged 495 of the 35,853 CVEs published in the first half of 2026
A post at The Hacker News puts published CVEs up roughly 49 percent year over year and observed exploitation at 495. For operators the figure that changes triage is the 116 attacked on the day they went public.
The Watch · Security desk
What happened
- A post at The Hacker News puts published CVEs for the first half of 2026 at 35,853, roughly 49 percent above the same period a year earlier.
- Of that total, 495 were catalogued as exploited in the wild during the same six months, about 1.4 percent of everything published.
- Anthropic disclosure data cited in the post shows Mythos-class models surfacing 26,153 vulnerability candidates in open-source software, with 421 patched upstream.
Compiled by The WatchSomething wrong?How this is made
Why it matters
- decision Roughly a quarter of the exploited set arrived with no patch window to schedule into, so the choice on those CVEs is which compensating control to move.
- constraint The validation-first model the post prescribes runs into its own coverage ceiling: at 32 percent of the attack surface a year, most assets are unvalidated at any given moment and a full pass takes over three years.
- contradiction The 1.4 percent figure counts confirmed observations, so it is a floor on real exploitation, and the post leans on it as though it were a measured share.
- exposure Air-gapped, restricted and business-critical assets are exactly the ones live exploit testing cannot touch. That leaves the highest-value estate short of the evidence the argument depends on.
Divide 35,853 by the 181 days in the half and the publication rate is 198 CVEs a day [1][4]. The catalogued exploitation rate is 2.7 a day [3][4]. Back the 49 percent increase out and the same six months of 2025 produced about 24,100, so roughly 11,800 more CVEs landed this year than last [2][3].
The figures come from a post at The Hacker News that argues for a single platform combining exploitability validation, security control validation and agentic pentesting [10]. It attributes the 26,153 candidate findings and 421 upstream patches to Anthropic's own disclosure data [5], and the pentest coverage numbers to Omdia [6]. The post does not name a source for the CVE and exploitation counts [7].
The 1.4 percent is a floor [1]. Catalogued exploitation is what someone observed, confirmed and published, and it is assigned after the fact.
116 of the 495 were already under attack when they became public [4]. That is about 23 percent of the exploited set, disclosure and use arriving together [2]. For those, compensating controls and reachability are the control.
The post's own numbers constrain its remedy. Omdia found 95 percent of organisations rank pentesting a top or high priority while only 32 percent of the average attack surface is tested each year [6]. That leaves 68 percent untested annually and puts one full pass at a little over three years [6]. The post also states the mechanical limits plainly: a working exploit has to exist, and business-critical, restricted and air-gapped assets may not be safe to run one against [9].
On the AI-discovery side, 421 of 26,153 candidates were patched upstream, about 1.6 percent, leaving 25,732 open [5][5][8]. Candidate volume of that size arriving at open-source maintainers is a supply problem for downstream users.
"Treating every vulnerability with a High or Critical CVSS rating as an emergency is not only impossible, it's actually the wrong model," the post said [7]. The scoring system rates a vulnerability. The same CVE can affect hundreds of assets and some of those instances are unreachable, while others sit behind controls that interrupt the technique [11]. What the 495 supports is demoting severity as the sort order for internal remediation work. Which 1.4 percent of next quarter's CVEs will be used is beyond it, and the post does not claim otherwise.
What to watch
- Whether a named database or exploitation catalogue is published behind the 35,853 and 495 counts. That would let the 1.4 percent ratio be checked.
- Whether the 116 same-day figure rises in the second half of 2026. A rise would further shorten patch-window planning.
- Whether Anthropic's disclosure data shows the 421 upstream patch count moving against the 26,153 candidates.