CVE-2026-82533
DeepSeek Harness < 0.1.2-alpha.1 Authentication Bypass via Host Header Spoofing
No known exploitation. EPSS puts it in the 69th percentile. Fixed in 0.1.2-alpha.1.
No public exploit knownFix0.1.2-alpha.1Upgrade to this build
No ready-made check. Neither Nuclei nor OpenVAS has one for this CVE yet.
What to do
Find your product and version, then upgrade to the build on the right.
- DeepSeek HarnessAffected: < 0.1.2-alpha.1Upgrade to: 0.1.2-alpha.1Stated by: CVE record
Versions as the sources state them. Full version matrix after sign-in
Steps
Written by AI from the record- Check whether your installed DeepSeek Harness version is older than 0.1.2-alpha.1.
- Check whether the Harness web service is reachable from outside your network (directly, via a tunnel/SSH forward, reverse proxy, or similar).
- If you’re using a vulnerable version, upgrade DeepSeek Harness to 0.1.2-alpha.1.
- After upgrading, restart the service and re-verify the running version matches 0.1.2-alpha.1.
Coverage timeline
2 publishersDeepSeek patched CVE-2026-82533, an authentication bypass in DeepSeek Harness that allowed AI agents to disable their file sandbox and execute commands without approval.
OX Research disclosed a critical vulnerability in DeepSeek Harness that allowed sandboxed AI agents to escape confinement via an unauthenticated local API, which was patched in version 0.1.2-alpha.1.
Sourced from public security reporting — each date is when it was published, not when it happened. 1 further article named this CVE in wider roundups without reporting it as a distinct event.
What it is
From the CVE record
DeepSeek Harness before 0.1.2-alpha.1 contains an authentication bypass vulnerability that grants unauthenticated access to its local HTTP agent-control API by accepting a client-supplied loopback Host header in place of validating the actual TCP connection origin. On the default configuration, a confined tool-executed process can reach the loopback API without any port exposure and use it to escape its own OS sandbox, escalate to unconfined execution, and disable the approval prompt. When the port is externally reachable via tunnel, SSH forward, or reverse proxy, a remote attacker can exploit the same flaw to create sessions, execute arbitrary commands, and exfiltrate stored conversation transcripts without credentials.
In plain language
Written by AI from the recordDeepSeek Harness has a critical login bypass where attackers can get into restricted features without credentials by spoofing a network “Host” value; this is a serious problem if your Harness service is reachable over the network in its default setup.
In DeepSeek Harness (before 0.1.2-alpha.1), an authentication bypass (CWE-807) can be triggered by Host header spoofing, letting an unauthenticated attacker reach and use restricted functions by the application trusting the client-supplied Host rather than the real connection source; the issue is reachable in default configuration.
If you're affected
- Unauthorized access to restricted features
- Stored data exposure (transcripts)
- System takeover risk
- Service disruption
Exploitation
Where each signal puts this CVE on the scale from published to confirmed exploited.
- CISA KEV
Not in the catalog. CISA has not confirmed exploitation.
- Public exploits
No public exploit or proof of concept found in the sources we track.
- EPSS
1.2% chance of exploitation activity in the next 30 days, which ranks it in the 69th percentile of scored CVEs.
Exploit Prediction Scoring System, FIRST.org. A probability, not a confirmation.
Lifecycle
4 events over 2 days, from the signal feeds we watch.
- Record updated
- Publishedweakness classified, record updated
Affected products
Technical detail
CVSS 3.1 vector
Open in the CVSS calculatorCVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:H/A:H
Scored 9.6 by NVD.
How it is reached
- Attack Vector NetworkExploitable remotely over the network without any special conditions
- Attack Complexity LowNo special conditions — the attack can be reliably reproduced
- Privileges Required NoneNo authentication required — anyone can exploit this
- User Interaction RequiredA user must click a link, open a file, or perform some action
Scope
- Scope ChangedThe exploit can affect other components (e.g. sandbox escape, host compromise from VM)
Impact if exploited
- Confidentiality HighTotal information disclosure — all data in the component is compromised
- Integrity HighTotal loss of integrity — attacker can modify any data in the component
- Availability HighTotal denial of service — the component is completely unavailable
Weaknesses
Sources
References in the record
- ox.security/blog/cve-2026-82533-deepseek-harness-ai-agent-sandbox-escape/
- github.com/deepseek-ai/deepseek-harness/releases/tag/dsh-v0.1.2-alpha.1
- github.com/deepseek-ai/deepseek-harness/commit/3e24087bfaeabe40b58ba2f7b936895b8f93fe27
And 1 more reference. See all after sign-in
In the news
Watch the software you run.
My Stack ranks new CVEs for your products by real-world exploitation, so the next serious one reaches you without reading every advisory.
We'll flag the next CVE, public exploit or patch for DeepSeek Harness, not every advisory.
A free account adds
- The full version matrix and every affected product
- Exploit links, proofs of concept and Metasploit modules
- Email alerts for the products you watch
- The same data over REST API, MCP and CLI