AI Agent Finds Dangerous Java Flaw Java4Shell in Three Hours
The startup Calif found a Java vulnerability called Java4Shell. According to the company, finding it and building a working attack took three hours and $200.
Calif, a cybersecurity startup, has found a critical flaw in Java, one of the most widely used programming languages. The company calls it Java4Shell. The bug is in the Base64 decoder, which turns plain text back into binary data. It can expose memory and let an attacker run their own code. Calif says finding it and building a working attack took three hours and $200 in AI usage fees. The work was done by OpenAI's coding agent, Codex.
Java powers more than 90% of the world's largest companies, including many government systems. Oracle, which maintains Java, has released a fix. Administrators should check that it is installed. There is no universal attack, because the risk depends on how each organization uses Base64.
The figures of three hours and $200 come from Calif alone. Nobody has checked them independently. The company did not name the AI model it used. The New York Times reported that it was GPT-6 Cyber. The researcher himself called much of the agent's output "slops." How many real findings such an agent produces, beyond this one, is not known.
Source: www.bankinfosecurity.com
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →▌ Related guides

Author
Evgenii Arsentev
PhD · Founder & CEO, Dad4Kids
Articles · Latest articles