CVE-2026-34452: CWE-59: Improper Link Resolution Before File Access ('Link Following') in anthropics anthropic-sdk-python
CVE-2026-34452 is a medium severity vulnerability in the anthropic-sdk-python versions 0. 86. 0 up to but not including 0. 87. 0. It involves improper link resolution before file access in the async local filesystem memory tool, allowing a local attacker with write access to the memory directory to exploit a race condition by replacing a symlink after path validation but before file operation. This can lead to sandbox escape by reading or writing files outside the intended directory. The synchronous memory tool is unaffected. The vulnerability requires local access with low privileges and high attack complexity, and no user interaction is needed. It has been patched in version 0.
AI Analysis
Technical Summary
The vulnerability CVE-2026-34452 affects the anthropic-sdk-python package, specifically versions from 0.86.0 up to but not including 0.87.0. The issue lies in the async local filesystem memory tool, which is designed to sandbox file operations within a designated memory directory. The tool performs validation to ensure that model-supplied file paths resolve inside the sandbox directory. However, after this validation step, it returns the unresolved path for subsequent file operations. This creates a time-of-check to time-of-use (TOCTOU) race condition where a local attacker with write permissions to the memory directory can replace or retarget a symbolic link (symlink) between the validation and the actual file operation. By doing so, the attacker can cause the program to read from or write to files outside the sandbox, effectively escaping the intended filesystem restrictions. This vulnerability is classified under CWE-59 (Improper Link Resolution Before File Access) and CWE-367 (Time-of-check Time-of-use Race Condition). The synchronous memory tool implementation in the SDK is not affected. The vulnerability requires local access with low privileges, has high attack complexity, and does not require user interaction. The CVSS 4.0 base score is 5.8, reflecting medium severity. The issue was patched in version 0.87.0 of the anthropic-sdk-python package. There are no known exploits reported in the wild at this time.
Potential Impact
This vulnerability allows a local attacker with write access to the memory directory used by the vulnerable async memory tool to escape the sandbox and perform unauthorized file reads or writes outside the intended directory. This could lead to unauthorized data disclosure, data tampering, or corruption of files on the host system. For organizations using the affected versions of anthropic-sdk-python in environments where untrusted users or processes have local write access to the memory directory, this could compromise the integrity and confidentiality of sensitive data. The impact is limited by the requirement for local access and the high complexity of the attack, but in multi-tenant or shared environments, or where the SDK is used in automated workflows with less strict access controls, the risk increases. Exploitation could undermine trust in sandboxing mechanisms and potentially facilitate privilege escalation or lateral movement if combined with other vulnerabilities.
Mitigation Recommendations
Organizations should upgrade anthropic-sdk-python to version 0.87.0 or later, where the vulnerability has been patched. Until upgrading is possible, restrict write permissions to the memory directory used by the async local filesystem memory tool to trusted users only, preventing untrusted local users from creating or modifying symlinks. Implement monitoring and alerting for unexpected symlink creation or modification in the memory directory. Consider using file system features or mandatory access controls (e.g., SELinux, AppArmor) to enforce strict sandbox boundaries. Review and audit local access controls on systems running the SDK to ensure that only authorized users have write access to relevant directories. Additionally, avoid running the vulnerable async memory tool in environments where local untrusted users exist. Finally, conduct thorough testing after patching to confirm that sandbox escapes are no longer possible.
Affected Countries
United States, United Kingdom, Germany, France, Japan, South Korea, Canada, Australia, India, Netherlands
CVE-2026-34452: CWE-59: Improper Link Resolution Before File Access ('Link Following') in anthropics anthropic-sdk-python
Description
CVE-2026-34452 is a medium severity vulnerability in the anthropic-sdk-python versions 0. 86. 0 up to but not including 0. 87. 0. It involves improper link resolution before file access in the async local filesystem memory tool, allowing a local attacker with write access to the memory directory to exploit a race condition by replacing a symlink after path validation but before file operation. This can lead to sandbox escape by reading or writing files outside the intended directory. The synchronous memory tool is unaffected. The vulnerability requires local access with low privileges and high attack complexity, and no user interaction is needed. It has been patched in version 0.
AI-Powered Analysis
Machine-generated threat intelligence
Technical Analysis
The vulnerability CVE-2026-34452 affects the anthropic-sdk-python package, specifically versions from 0.86.0 up to but not including 0.87.0. The issue lies in the async local filesystem memory tool, which is designed to sandbox file operations within a designated memory directory. The tool performs validation to ensure that model-supplied file paths resolve inside the sandbox directory. However, after this validation step, it returns the unresolved path for subsequent file operations. This creates a time-of-check to time-of-use (TOCTOU) race condition where a local attacker with write permissions to the memory directory can replace or retarget a symbolic link (symlink) between the validation and the actual file operation. By doing so, the attacker can cause the program to read from or write to files outside the sandbox, effectively escaping the intended filesystem restrictions. This vulnerability is classified under CWE-59 (Improper Link Resolution Before File Access) and CWE-367 (Time-of-check Time-of-use Race Condition). The synchronous memory tool implementation in the SDK is not affected. The vulnerability requires local access with low privileges, has high attack complexity, and does not require user interaction. The CVSS 4.0 base score is 5.8, reflecting medium severity. The issue was patched in version 0.87.0 of the anthropic-sdk-python package. There are no known exploits reported in the wild at this time.
Potential Impact
This vulnerability allows a local attacker with write access to the memory directory used by the vulnerable async memory tool to escape the sandbox and perform unauthorized file reads or writes outside the intended directory. This could lead to unauthorized data disclosure, data tampering, or corruption of files on the host system. For organizations using the affected versions of anthropic-sdk-python in environments where untrusted users or processes have local write access to the memory directory, this could compromise the integrity and confidentiality of sensitive data. The impact is limited by the requirement for local access and the high complexity of the attack, but in multi-tenant or shared environments, or where the SDK is used in automated workflows with less strict access controls, the risk increases. Exploitation could undermine trust in sandboxing mechanisms and potentially facilitate privilege escalation or lateral movement if combined with other vulnerabilities.
Mitigation Recommendations
Organizations should upgrade anthropic-sdk-python to version 0.87.0 or later, where the vulnerability has been patched. Until upgrading is possible, restrict write permissions to the memory directory used by the async local filesystem memory tool to trusted users only, preventing untrusted local users from creating or modifying symlinks. Implement monitoring and alerting for unexpected symlink creation or modification in the memory directory. Consider using file system features or mandatory access controls (e.g., SELinux, AppArmor) to enforce strict sandbox boundaries. Review and audit local access controls on systems running the SDK to ensure that only authorized users have write access to relevant directories. Additionally, avoid running the vulnerable async memory tool in environments where local untrusted users exist. Finally, conduct thorough testing after patching to confirm that sandbox escapes are no longer possible.
Technical Details
- Data Version
- 5.2
- Assigner Short Name
- GitHub_M
- Date Reserved
- 2026-03-27T18:18:14.895Z
- Cvss Version
- 4.0
- State
- PUBLISHED
Threat ID: 69cc424fe6bfc5ba1d44f4b9
Added to database: 3/31/2026, 9:53:19 PM
Last enriched: 3/31/2026, 10:09:22 PM
Last updated: 3/31/2026, 11:19:26 PM
Views: 2
Community Reviews
0 reviewsCrowdsource mitigation strategies, share intel context, and vote on the most helpful responses. Sign in to add your voice and help keep defenders ahead.
Want to contribute mitigation steps or threat intel context? Sign in or create an account to join the community discussion.
Actions
Updates to AI analysis require Pro Console access. Upgrade inside Console → Billing.
Need more coverage?
Upgrade to Pro Console for AI refresh and higher limits.
For incident response and remediation, OffSeq services can help resolve threats faster.
Latest Threats
Check if your credentials are on the dark web
Instant breach scanning across billions of leaked records. Free tier available.