CVE-2026-55437: Anthropic Python stored XSS

CISO Take

A stored cross-site scripting flaw in Coder's workspace dashboard let any user who could run a workspace embed live HTML into agent log output, which then rendered unsanitized in the browser of anyone else, including an administrator, who viewed that workspace's logs. The blast radius is meaningful given Coder's 5,435 downstream dependents, but real-world risk is tempered by the fact that exploitation demands both workspace-owner privileges and a victim actively opening the affected log view, there is no EPSS score, no CISA KEV listing, and no public exploit or Nuclei template observed. CSP blocked the most dangerous outcome (inline script execution), leaving the practical impact limited to phishing-style meta-refresh redirects, CSS-based UI redressing, and passive data exfiltration via external image beacons rather than full session takeover. Patch to 2.34.2, 2.33.8, 2.32.7, or 2.29.17 depending on your release line; there is no workaround, so treat this as a standard patch-cycle item rather than an emergency, prioritized higher if your Coder instance is used for AI agent development workspaces with admin-level log visibility.

Sources: GitHub Advisory CISA KEV ATLAS

What is the risk?

CVSS 5.4 (medium) reflects a scope-changed vulnerability (S:C) with low complexity but two mitigating gates: privileges required (workspace-owner access, PR:L) and user interaction (a victim must open the workspace page, UI:R). Confidentiality and integrity impact are both rated low and availability is unaffected, consistent with a UI-rendering flaw rather than code execution — CSP's block on inline scripts removes the highest-severity outcome. No EPSS score, no CISA KEV entry, and no public PoC or scanner template exist, indicating low current exploitation likelihood. The residual risk is concentrated in environments where less-trusted users (e.g., external contractors, shared AI dev sandboxes) can own workspaces while administrators routinely review agent logs, since the attack specifically targets that trust boundary.

How does the attack unfold?

Malicious log injection

A user with workspace-owner privileges crafts agent output containing unescaped HTML metacharacters (e.g., a meta-refresh tag or style block).

AML.T0008.000

Victim views dashboard

An administrator or other user opens the workspace page in the Coder dashboard to review agent logs.

Unsanitized rendering

The AgentLogLine component inserts the attacker's HTML via dangerouslySetInnerHTML without escaping, causing it to render as live markup in the victim's session.

AML.T0011.003

UI manipulation or redirect

Blocked from running inline scripts by CSP, the payload instead redirects the victim via meta-refresh, applies deceptive CSS overlays, or triggers external image requests to exfiltrate context.

AML.T0073

Malicious log injection

A user with workspace-owner privileges crafts agent output containing unescaped HTML metacharacters (e.g., a meta-refresh tag or style block).

AML.T0008.000

Victim views dashboard

An administrator or other user opens the workspace page in the Coder dashboard to review agent logs.

Unsanitized rendering

The AgentLogLine component inserts the attacker's HTML via dangerouslySetInnerHTML without escaping, causing it to render as live markup in the victim's session.

AML.T0011.003

UI manipulation or redirect

Blocked from running inline scripts by CSP, the payload instead redirects the victim via meta-refresh, applies deceptive CSS overlays, or triggers external image requests to exfiltrate context.

AML.T0073

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
Anthropic Python	go	>= 2.34.0, < 2.34.2	`2.34.2`
3.8K 5.2K dependents Pushed yesterday 90% patched ~11d to patch Full package profile →

Do you use Anthropic Python? You're affected.

How severe is it?

CVSS 3.1

5.4 / 10

EPSS

0.2%

chance of exploitation in 30 days

Higher than 8% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

No known exploitation

Sophistication

Trivial

What is the attack surface?

AV Network

AC Low

PR Low

UI Required

S Changed

C Low

I Low

A None

What should I do?

1 step

Upgrade immediately to the patched release for your line: 2.34.2, 2.33.8, 2.32.7, or 2.29.17 (ESR) — the fix simply enables escapeXML: true in the ansi-to-html conversion before DOM insertion. There is no workaround, so patching is the only remediation path. In the interim, restrict who can create or own workspaces to trusted users, and be cautious when reviewing agent logs from workspaces owned by lower-trust accounts, especially as an administrator. For detection, review CSP violation reports for blocked inline-script attempts on workspace/log pages, and audit stored agent log content for embedded <meta>, <style>, or <img> tags with external URLs, which would indicate attempted exploitation.

What does CISA's SSVC say?

Decision Track

Exploitation none

Automatable No

Technical Impact partial

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

Social Engineering Data Leakage Agent AML.T0008.000 - AI Development Workspaces AML.T0011.003 - Malicious Link

Which compliance frameworks are affected?

This CVE is relevant to:

ISO 42001

A.6.2 - AI system development environment security

NIST AI RMF

MANAGE 4.1 - Third-party AI system component risk management

Frequently Asked Questions

What is CVE-2026-55437?

A stored cross-site scripting flaw in Coder's workspace dashboard let any user who could run a workspace embed live HTML into agent log output, which then rendered unsanitized in the browser of anyone else, including an administrator, who viewed that workspace's logs. The blast radius is meaningful given Coder's 5,435 downstream dependents, but real-world risk is tempered by the fact that exploitation demands both workspace-owner privileges and a victim actively opening the affected log view, there is no EPSS score, no CISA KEV listing, and no public exploit or Nuclei template observed. CSP blocked the most dangerous outcome (inline script execution), leaving the practical impact limited to phishing-style meta-refresh redirects, CSS-based UI redressing, and passive data exfiltration via external image beacons rather than full session takeover. Patch to 2.34.2, 2.33.8, 2.32.7, or 2.29.17 depending on your release line; there is no workaround, so treat this as a standard patch-cycle item rather than an emergency, prioritized higher if your Coder instance is used for AI agent development workspaces with admin-level log visibility.

Is CVE-2026-55437 actively exploited?

No confirmed active exploitation of CVE-2026-55437 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-55437?

Upgrade immediately to the patched release for your line: 2.34.2, 2.33.8, 2.32.7, or 2.29.17 (ESR) — the fix simply enables `escapeXML: true` in the ansi-to-html conversion before DOM insertion. There is no workaround, so patching is the only remediation path. In the interim, restrict who can create or own workspaces to trusted users, and be cautious when reviewing agent logs from workspaces owned by lower-trust accounts, especially as an administrator. For detection, review CSP violation reports for blocked inline-script attempts on workspace/log pages, and audit stored agent log content for embedded `<meta>`, `<style>`, or `<img>` tags with external URLs, which would indicate attempted exploitation.

What systems are affected by CVE-2026-55437?

This vulnerability affects the following AI/ML architecture patterns: agent frameworks, AI development workspaces, remote development environments.

What is the CVSS score for CVE-2026-55437?

CVE-2026-55437 has a CVSS v3.1 base score of 5.4 (MEDIUM). The EPSS exploitation probability is 0.18%.

What is the AI security impact?

Affected AI Architectures

agent frameworksAI development workspacesremote development environments

MITRE ATLAS Techniques

AML.T0008.000 AI Development Workspaces

AML.T0011.003 Malicious Link

Compliance Controls Affected

ISO 42001: A.6.2

NIST AI RMF: MANAGE 4.1

What are the technical details?

Original Advisory

### Summary The `AgentLogLine` dashboard component instantiated `ansi-to-html` without `escapeXML: true` and inserted the result via `dangerouslySetInnerHTML` so HTML embedded in workspace agent log lines was rendered as live markup. Server-side sanitization did not neutralize HTML metacharacters. > **Note:** Exploitation requires a victim to view attacker-controlled agent logs in the dashboard. ### Impact A user who could run a workspace could emit arbitrary HTML into agent logs; when another user, including an administrator, viewed the workspace page, it rendered in their session. Content Security Policy blocked inline scripts but an attacker could still inject a `meta refresh` redirect, `style` rules for UI redressing or CSS-based exfiltration or external `img` beacons. This required workspace-owner access and a victim viewing the page. ### Patches The fix enables `escapeXML: true` so HTML metacharacters are escaped before DOM insertion. The fix was backported to all supported release lines: | Release line | Patched version | |---|---| | 2.34 | [v2.34.2](https://github.com/coder/coder/releases/tag/v2.34.2) | | 2.33 | [v2.33.8](https://github.com/coder/coder/releases/tag/v2.33.8) | | 2.32 | [v2.32.7](https://github.com/coder/coder/releases/tag/v2.32.7) | | 2.29 (ESR) | [v2.29.17](https://github.com/coder/coder/releases/tag/v2.29.17) | ### Workarounds None. ### Resources - Fix: #25808 ### Credits Coder would like to thank Anthropic's Security Team (ANT-2026-22449) for independently disclosing this issue!

Exploitation Scenario

An attacker who has been granted the ability to run a workspace (e.g., a contractor, intern, or compromised low-privilege account in an AI dev environment) crafts a startup script or process whose output includes HTML metacharacters, such as `<meta http-equiv="refresh" content="0;url=https://attacker.example/login">` or `<style>` rules designed for UI redressing. This text is written to the workspace's agent log. When a higher-privileged user, most impactfully an administrator, later opens that workspace's dashboard page to inspect logs, the unsanitized HTML is inserted via `dangerouslySetInnerHTML` and renders live: the admin's browser may be silently redirected to a credential-harvesting page styled to mimic Coder's login, or CSS-based techniques may overlay deceptive UI elements or trigger external image requests that beacon session context to the attacker — all without needing to execute JavaScript, since CSP blocks that vector.

Weaknesses (CWE)

CWE-79 Improper Neutralization of Input During Web Page Generation ('Cross-site Scripting') Primary CWE-79 Improper Neutralization of Input During Web Page Generation ('Cross-site Scripting')

CWE-79 — Improper Neutralization of Input During Web Page Generation ('Cross-site Scripting'): The product does not neutralize or incorrectly neutralizes user-controllable input before it is placed in output that is used as a web page that is served to other users.

[Architecture and Design] Use a vetted library or framework that does not allow this weakness to occur or provides constructs that make this weakness easier to avoid [REF-1482]. Examples of libraries and frameworks that make it easier to generate properly encoded output include Microsoft's Anti-XSS library, the OWASP ESAPI Encoding module, and Apache Wicket.
[Implementation, Architecture and Design] Understand the context in which your data will be used and the encoding that will be expected. This is especially important when transmitting data between different components, or when generating outputs that can contain multiple encodings at the same time, such as web pages or multi-part mail messages. Study all expected communication protocols and data representations to determine the required encoding strategies. For any data that will be output to another web page, especially any data that was received from external inputs, use the appropriate encoding on all non-alphanumeric characters. Parts of the same output document may require different encodings, which will vary depending on whether the output is in the: etc. Note that HTML Entity Encoding is only appropriate for the HTML body. Consult the XSS Prevention Cheat Sheet [REF-724] for more details on the types of encoding and escaping that are needed. HTML body Element attributes (such as src="XYZ") URIs JavaScript sections Casca

Source: MITRE CWE corpus.