/

CVE-2026-92785

Severity:

HIGH RISK

(8.1)

Unsafe Deserialization (CWE-502) in Angel Master RPC

CVE-2026-92785

Deferred

Vulnerability type

Category:

Unsafe Deserialization

CWE:

CWE-502

Impact:

Arbitrary Class Instantiation and Denial of Service

Severity:

HIGH RISK

(8.1)

CVSS version:

3.1

Catch this CVE while code is being written

Detect vulnerable code directly in pull requests, so CVEs are fixed before merge.

Summary

Angel through version 3.3.0 deserializes untrusted setAlgoMetrics payloads with Kryo without class registration or allowlist validation. An unauthenticated network attacker may instantiate unintended classes or exhaust coordinator memory by sending crafted serialized objects to the master RPC endpoint.

Affected context

The vulnerability occurs in Angel deployments where the master RPC endpoint accepts setAlgoMetrics requests and processes their serialized payloads with Kryo. Risk is increased when the endpoint is network-reachable by untrusted clients and deserialization is performed without class registration, an allowlist, or equivalent input constraints.

Exploit preconditions

  • Attack surface: Angel master RPC endpoint
  • Required access: Network connectivity to the endpoint; authentication is not required according to the vulnerability description
  • Trigger point: Processing of a crafted serialized setAlgoMetrics payload
  • Why exploitation is possible: Kryo deserializes attacker-controlled object data without restricting permitted classes, allowing unintended object construction or resource-intensive processing

Why this happens

The underlying bug is unsafe deserialization: attacker-controlled serialized data is treated as a trusted object graph. Kryo is used without class registration or an allowlist that limits the types eligible for reconstruction. This insecure pattern maps to CWE-502 because deserialization can invoke object construction behavior and consume resources before the input has been adequately constrained.

Potential impact

  • An attacker may instantiate unintended classes within the Angel coordinator process.
  • The coordinator may consume excessive memory and become unavailable.
  • The master service could experience degraded performance or denial of service.
  • Unexpected object behavior could affect application integrity or operational stability.

General mitigation guidance

Mitigation generally focuses on removing trust from serialized RPC inputs and

restricting the object types that can be reconstructed.

Common defensive patterns include:

  • Prefer explicit, schema-driven data formats and simple data-transfer objects over native object deserialization.
  • Enforce strict class registration or a narrowly defined allowlist when Kryo or a similar serializer is required.
  • Validate payload size, nesting depth, collection limits, and field values before resource-intensive processing.
  • Require authentication and authorization for administrative or coordinator RPC operations.
  • Run coordinator services with least privilege and apply resource limits to reduce denial-of-service impact.
  • Keep RPC endpoints isolated from untrusted networks where practical.

How teams detect issues like this

Organizations can look for this weakness through:

  • Static analysis rules identifying deserialization of network-controlled data without registration or allowlist checks.
  • Code review signals such as generic Kryo readers, native object reconstruction, and missing payload limits.
  • RPC inventory reviews that identify unauthenticated endpoints accepting serialized object graphs.
  • Runtime monitoring for unusual setAlgoMetrics request rates, oversized payloads, memory growth, and coordinator restarts.
  • Policy enforcement requiring schema validation, type restrictions, authentication, and resource bounds for RPC services.

Automated code review and software composition analysis tools may help identify related patterns, but should be supplemented with service-level testing.

Sources

SourceLink
github.comgithub.com/Angel-ML/angel
github.comgithub.com/Angel-ML/angel/blob/Release-3.3.0/angel-ps/core/src/main/java/com/tencent/angel/utils/KryoUtils.java#L45-L70
github.comgithub.com/Angel-ML/angel/issues/1355
vulncheck.comvulncheck.com/advisories/angel-through-3.3.0-unauthenticated-kryo-deserialization-of-arbitrary-classes

Frequently asked

questions

Yes. The described attack surface is the master RPC endpoint, and exploitation

may be possible from an unauthenticated network position when that endpoint is

reachable.

The attacker needs network connectivity to the relevant Angel master RPC

service. The description does not require prior authentication, although

network segmentation and gateway controls may limit reachability.

Avoid deserializing native object graphs from untrusted RPC input. Use a

schema-driven format or enforce strict Kryo class registration, allowlists,

payload limits, and validation before object processing.

Require authentication and authorization for sensitive RPC methods, restrict

endpoint exposure, apply resource limits, and use code review and automated

policy checks to detect unsafe deserialization patterns.

Find out what's already
exploitable in your codebase.