Someone reported java4shell bug in September 2026
An overlooked OpenJDK pull request from 2026 that clearly described a Base64-related heap corruption is now being linked to the recent “java4shell” vulnerability, raising questions about how security bugs in critical infrastructure software are triaged and communicated. Commenters debate whether Oracle handled the report responsibly under strict embargo-style policies, or whether auto-close bots, opaque processes, and fear of legal risk led to a month-long delay on a publicly visible zero-day. The thread also highlights how large language models may both uncover and quietly reuse prior work on vulnerabilities, compressing disclosure timelines and challenging traditional responsible-disclosure practices.
Handling of the original Java buffer overflow PR
- A buffer overflow in a JDK Base64 intrinsic was publicly filed as a PR in September, clearly described as heap corruption.
- Some commenters argue it should have been treated as a confidential security report from the start, not a public PR.
- Others say it was initially seen as “just a crash” rather than clear RCE, which often leads to public handling.
- Evidence from a related issue shows Oracle privately asked the reporter not to proceed with a PR, then “took over” the fix; the reporter later said nothing else was required from them.
OpenJDK / Oracle security process
- Several participants claim that for JDK, any VM crash or compiler backend issue is flagged to a security team; once that happens, public communication and merges stop until triage is complete.
- Some defend this as necessary for infrastructure-critical software with coordinated quarterly releases.
- Critics counter that once a vulnerability is public (e.g., via a PR), a month+ to fix is too slow; silence doesn’t “put the cat back in the bag.”
- There’s debate over whether they should have immediately disabled the intrinsic and shipped a fast CVE vs waiting for full risk assessment.
Auto-close bots and project triage
- Many blame “stale” bots and lack of human triage for the PR’s neglect, calling auto-close a sign of bad project management.
- Others note the PR was closed by the author after a bot warning, and the real problem is volume and limited maintainer time.
- One JDK contributor states bugs “don’t slip triage” and suggests the quiet was due to deliberate security handling, not neglect.
Role of LLMs and simultaneous discovery
- Some suspect the later AI-assisted research might have surfaced or reused the earlier public PR; others call that speculative or unnecessary.
- Multiple reports (fuzzing, independent analysis) suggest genuine simultaneous discovery is common, especially now.
- There’s debate over whether LLMs are mere tools (like compilers) or primary agents when a simple prompt yields a critical find.
- Participants predict LLM-based triage on PRs/issues will become standard and that AI-driven vuln finding will compress acceptable disclosure timelines.