Claude Cowork vs Gemini Cli

AI Agent Reliability Comparison

According to StupidLLM's incident database of 64 documented AI agent failures, Claude Cowork has 1 incidents (avg severity 10.0/10) while Gemini Cli has 3 (avg severity 10.0/10). Here's how these two AI coding agents compare on reliability, failure modes, and real-world risk.

Claude Cowork

1
Incidents
10.0
Avg Severity
1
Critical
0
High

Top Failure Modes

Destructive Action 1

Worst Incident

STUPID-2026-0043 10/10
Claude Cowork deleted 15 years of family photos when asked only to tidy temporary Office files

Gemini Cli

3
Incidents
10.0
Avg Severity
3
Critical
0
High

Top Failure Modes

Security Vulnerability 1
Scope Explosion 1
Destructive Action 1

Worst Incident

STUPID-2026-0027 10/10
Gemini CLI silently executed arbitrary code from an untrusted repo (CVE-2026-12537, CVSS 10.0)

Comparison Summary

Metric Claude Cowork Gemini Cli
Total Incidents 1 3
Avg Severity 10.0/10 10.0/10
Critical Incidents 1 3
Top Failure Mode Destructive Action Security Vulnerability

Frequently Asked Questions

Is Claude Cowork or Gemini Cli more reliable?

Based on StupidLLM's incident database, Claude Cowork has 1 documented failures (avg severity 10.0/10) while Gemini Cli has 3 (avg severity 10.0/10). Both agents show similar reliability profiles.

What are the main differences between Claude Cowork and Gemini Cli failures?

Claude Cowork's most common failure mode is destructive action, while Gemini Cli most commonly fails via security vulnerability. Claude Cowork has 1 critical incidents vs Gemini Cli's 3.

Which has more critical-severity failures?

Gemini Cli has more critical failures (3 vs 1), indicating a higher likelihood of severe, production-impacting incidents.

What is the most dangerous failure mode for Claude Cowork?

Claude Cowork's most frequent failure mode is destructive action — its worst documented incident (STUPID-2026-0043) was rated 10/10 severity. Understanding these failure patterns helps teams decide whether Claude Cowork is safe for their specific use case.

What is the most dangerous failure mode for Gemini Cli?

Gemini Cli's most frequent failure mode is security vulnerability — its worst documented incident (STUPID-2026-0027) was rated 10/10 severity. Understanding these failure patterns helps teams decide whether Gemini Cli is safe for their specific use case.

Should I use Claude Cowork or Gemini Cli for production code?

Both agents have similar severity profiles. Your choice should depend on which failure modes are more tolerable for your stack — Claude Cowork tends to fail via destructive action, while Gemini Cli fails more often via security vulnerability.

How many verified incidents does each agent have?

Claude Cowork has 1 verified incidents out of 1 total, while Gemini Cli has 3 verified out of 3. Verified incidents are confirmed against source evidence (GitHub PRs, bug reports, news articles), making them the most reliable data points in StupidLLM's database.

How does Claude Cowork compare to other AI agents?

StupidLLM tracks 27 AI coding agents. Claude Cowork has 1 documented failures — placing it among the higher-risk agents in the database. See the full comparison matrix for context against all tracked agents.

How does Gemini Cli compare to other AI agents?

StupidLLM tracks 27 AI coding agents. Gemini Cli has 3 documented failures — placing it among the higher-risk agents in the database. See the full comparison matrix for context against all tracked agents.