Security1 publisher3 min readPublished
Progress fixed the chain on July 8, and TantoSec published the method plus a working tool on September 7, so the exposed installs are the ones where an administrator set the explicit upload encryption key Telerik recommends.
The Watch · Security desk

Compiled by The WatchSomething wrong?How this is made
RadAsyncUpload encrypts its client-side upload configuration with AES-CBC and no integrity check, so the server answers tampered ciphertext differently depending on whether the decrypted bytes have valid padding or merely fail to parse as JSON [11]. That difference is CVE-2026-13182, and it lets an attacker read the blob without the key; because the control derives its encryption from a fixed seed, TantoSec turned reading into writing and forged a configuration of its own [12]. A forged configuration can name any .NET type. The control resolves the name with no allowlist, which is CVE-2026-13181, and deserializes it into a gadget that loads a DLL from a location the attacker controls [13]. The DLL is a mixed-mode assembly, so native code runs the moment it loads [14]. Where the application suppresses detailed errors, the oracle is still readable through response timing, tracked separately as CVE-2026-13183 [16].
None of that is quiet. TantoSec's end-to-end run took roughly 127,000 oracle requests, about an hour against a lab target and longer against a rate-limited server [15]. An hour for 127,000 requests is about 35 per second sustained against a single upload handler [3]. Rate limiting stretches the clock rather than closing the path [15], which makes request volume the practical detection surface for anyone whose change window is weeks out.
Version scanning gets you part way. The affected range is 2010.1.309 through 2026.2.519, with 2026.2.708 and later fixed [5], roughly sixteen years of releases in scope [4]. What the version does not tell you is which of those applications render a RadAsyncUpload control whose server-side handler reads the upload result, and which have an explicit encryption key set for the control [7]. That second condition is the awkward one, because setting the key is hardening Telerik recommends [8]. An affected install missing either condition is not exploitable through this chain [9].
The tooling landed 61 days after the fix shipped and 47 days after Progress published the CVEs [1][2]. Attack-surface vendor IONIX says on its site that it is tracking ongoing exploitation attempts, but supplies no dates or volumes and does not separate exploitation from ordinary internet scanning of the handler [18]. Treat that as unconfirmed until someone puts a date on it.
The handler's history is with older bugs. CVE-2019-18935, a deserialization flaw in the same component, was chained with a 2017 encryption weakness and used by ransomware crews and nation-state actors, including in a 2022 breach of a U.S. federal agency, and was still being exploited in 2025 [19]. CVE-2026-13181 is scored 8.1 with high attack complexity, and that rating describes the configuration prerequisites, not any difficulty once they are met [6]. For most affected estates the work is a version bump. For the subset with an explicit upload key, the work is finding them before someone else finishes 127,000 requests.
Ranked by verification strength, evidence, and original report placement.
Security firm TantoSec published a working exploit chain against Telerik UI for ASP.NET AJAX that allows an unauthenticated attacker to execute remote code on the server hosting a vulnerable application.
On September 7, TantoSec's Marcio Almeida published a walk-through of the full chain plus a command-line tool, telerik-rau-exploit, and two mixed-mode DLL payloads: one that writes a web shell to disk and one that runs entirely in memory.
Progress Software shipped the fix in version 2026.2.708 (2026 Q2 SP1) on July 8.
Per Progress's advisory, the chain affects the RadAsyncUpload file-upload control in versions 2010.1.309 through 2026.2.519; 2026.2.708 and later are fixed.
CVE-2026-13181, an unguarded type-resolution flaw, carries a CVSS score of 8.1 (high); its high attack-complexity rating reflects the configuration prerequisites rather than any difficulty in exploitation once they are met.
Follow any of these and your For You feed starts watching them — no settings page required.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One newsroom, checkable primaries
Every technical figure in this story is attributed to something a reader can open: TantoSec's write-up for the mechanics and the 127,000-request cost, Progress's advisory for the version range and CVSS 8.1, CISA's catalog for the exploitation check. The ceiling is arithmetic rather than quality. Two copies of one Hacker News report is one newsroom, and nobody has re-run the chain independently or counted how many installs meet the preconditions.
Tooling public, use unobserved
What has actually happened is dated and modest: a fixed release on July 8, an advisory on July 22, and a working tool on September 7, while CISA's catalog still lists none of these CVEs as exploited. The population that decides whether any of this matters, installs rendering the control with a custom encryption key set, has no published number attached to it in this reporting.
Roughly aligned
The headline sells unauthenticated remote code execution and the body spends most of its length walking that back to a narrow configuration, which is the right order of operations. The one inflated note in circulation belongs to IONIX, and this reporting quotes it, then declines to treat it as evidence. Slight positive because a reader skimming the title alone would assume broader exposure than the two preconditions allow.
Interested parties on each side of the alarm
Every named party has a stake. TantoSec's tool release doubles as its shop window, IONIX sells exposure scanning and is the sole voice asserting exploitation, and Progress's single official remedy is to move onto its current release. None of that makes the technical account wrong, and the reporting names each interest openly, which is why the pressure is legible rather than buried.
Solid mechanics, unknown exposure
The mechanics hold up well: specific, internally consistent, and checkable against a vendor advisory and a published tool. Exposure and exploitation rest on weaker footing, though. The number of applications meeting both preconditions is unknown, the sole exploitation claim carries no dates, and everything arrives through a single publisher, so a correction anywhere in the chain would have nothing in this coverage to catch it.
security
UAT-10147 runs public exploits through an AI pipeline, and old CVEs are still the door2 publishers
security
Google patches a V8 type confusion already being exploited against Chrome users9 publishers
security
CISA ties federal patch deadlines to four yes-or-no questions about each CVE1 publisher
security
Oracle's monthly patch drop hit 925 CVEs in August, four times June's volume2 publishers
Publishers with included, body-backed reporting in this cluster.
2 articles · September 7, 2026