Is CVE-2026-95183 exploited in the wild?

No. Junglewise Threat Intelligence has no report of CVE-2026-95183 being exploited in the wild.

How severe is CVE-2026-95183?

Informational severity, published 16 October 2026.

Which products and versions are affected?

DeepSeek Harness (dsh) developer preview, commit c291e796 (2026-09).

Is there a fix for CVE-2026-95183?

No fixed version is recorded yet. Check the vendor's advisory in the references below. If you run it, Junglewise drafts a mitigation that fits your setup and applies it once you approve, with senior security engineers behind it.

How is CVE-2026-95183 attacked?

It is attacked over the network, in a software library.

Am I affected by CVE-2026-95183?

Check whether you run DeepSeek Harness (dsh) at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

CVE-2026-95183: DeepSeek Harness prompt injection via context-compaction

CVE-2026-95183 · Severity: info · Published 2026-10-16

DeepSeek Harness (dsh) is a desktop AI assistant that automatically summarizes conversation history to maintain context across sessions. An attacker can inject malicious instructions into external content that the assistant reads—such as web pages, emails, or files—and these instructions are preserved verbatim in the summary and executed as legitimate tasks in subsequent sessions, potentially leading to silent data exfiltration or unintended command execution.

Vendors: DeepSeek.

Quick answers

Is CVE-2026-95183 exploited in the wild?: No. Junglewise Threat Intelligence has no report of CVE-2026-95183 being exploited in the wild.
How severe is CVE-2026-95183?: Informational severity, published 16 October 2026.
Which products and versions are affected?: DeepSeek Harness (dsh) developer preview, commit c291e796 (2026-09).
Is there a fix for CVE-2026-95183?: No fixed version is recorded yet. Check the vendor's advisory in the references below. If you run it, Junglewise drafts a mitigation that fits your setup and applies it once you approve, with senior security engineers behind it.
How is CVE-2026-95183 attacked?: It is attacked over the network, in a software library.
Am I affected by CVE-2026-95183?: Check whether you run DeepSeek Harness (dsh) at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

Technical details

The vulnerability is an improper neutralization of LLM prompts (CWE-1427) in the context-compaction feature of the @deepseek-ai/dsh-compaction-basic component. When the model summarizes conversation history, it reproduces user requests and external tool outputs verbatim without distinguishing trusted user intent from untrusted injected directives. Attackers achieve code execution or data exfiltration by embedding shell commands or operational requirements into external content; these persist across compaction boundaries as established tasks with no user notification. Attack requires only network access to injectable content sources (AV:N, AC:L, PR:N, UI:N).

Affected products

DeepSeek Harness (dsh) developer preview, commit c291e796 (2026-09)

Timeline

2026-10-16: disclosed: CVE-2026-95183 published
2026-10-14: other: PoC published on GitHub

References

This page as Markdown. Free to reuse under CC BY 4.0 with a link back.