The flaw lies in the resolve_type function of letta-ai letta 0.16.4, where input is directly passed to eval without proper sanitization. This flaw allows attackers to execute arbitrary code through crafted inputs, potentially compromising the host system. The vulnerability corresponds to code injection weaknesses, specifically dynamic code execution (CWE-94) and unsafe eval usage (CWE-95).

The issue is present in the letta-ai letta application, version 0.16.4. No other versions are identified in the vendor data, and the vendor name is letta-ai with the product letta. The common platform enumeration indicates a generic application code base exposed to user input.

The CVSS base score is 6.9, placing the vulnerability in the medium to high risk range. EPSS data is not available, but publicly available exploit code confirms that the issue can be leveraged remotely. The attacker can trigger the flaw from outside the application, and because no mitigation was released by the vendor, the risk remains until the software is updated or alternative safeguards are applied. The vulnerability is not yet listed in the KEV catalog, suggesting it may not have widespread exploitation yet, yet the public exploit increases the likelihood of future attacks.