← AI Switchboard
SAFETY · November 19, 2026
Nov 17

Calif says OpenAI's Codex helped it find Java4Shell, a critical remote-code flaw in OpenJDK, in an hour for less than $200

By the AI Switchboard · researched and written with AI from the sources listed on this page, each point checked against its source before publishing · how we work · report a mistake

Oracle shipped fixes on 17 November. It is Calif's second AI-assisted find after the PHP flaws the Switchboard covered on 23 October.

The security firm Calif disclosed a flaw it calls Java4Shell on 17 November. It is a memory-corruption bug in the Base64 decoder that OpenJDK's HotSpot runs on x86-64 processors with AVX-512 VBMI instructions, and Calif says it can let an attacker run code remotely. Calif says it affects OpenJDK 17.0.6 and later, and the distributions built from it.

Calif's researcher Quang Luong pointed OpenAI's Codex at likely weak spots. “It took our own Quang Luong an hour to discover the vulnerability, and two more hours to develop the first working exploit,” Calif writes. “The total token cost was less than $200.” Calif rates the flaw 9.2 out of 10 under CVSS 4.0.

Two sources side by side: Calif's 9.2 is its own score under CVSS 4.0, while Oracle rates the flaw 8.1 under CVSS 3.1, according to Qualys. Oracle released patched Java builds on 17 November. Qualys wrote that as of that day it had seen no public confirmation of exploitation, and Cloudflare, which rolled out firewall detections on 18 November, said it had seen no sign of exploitation either.

What changed since the Switchboard's story of 23 October, when PHP patched two critical flaws found by AI-assisted Wiz researchers working with Calif and OpenAI: a month later, the same approach has found a flaw in Java, which runs much of the world's server software, and Calif puts a price on it.

  • Claimed Calif says Codex helped its researcher find the flaw in an hour and build a working exploit in two more, for less than $200 in tokens. Calif, Java4Shell write-up
  • Reported Oracle released fixes on 17 November and rates the flaw 8.1 under CVSS 3.1; Qualys had seen no public confirmation of exploitation as of 17 November. Qualys
  • Reported Cloudflare rolled out detections and said it saw no sign of exploitation. Cloudflare Radar

Safety, security & governanceAgents in the wild

Catching up in the November 19, 2026 edition · front page