OpenAI training models to write superhumanly secure code, says Brockman
tekbog · x · 2026-08-17
Greg Brockman says OpenAI is training its models to write "superhumanly secure code" as part of efforts to harden institutional infrastructure against AI-powered attacks.
The replying user jokes about adding "write superhumanly secure code" to all their prompts.
More from coding & agent
- Left the room, came back to find his laptop and watch having a voice chat about him — johnlindquist · 2026-08-17
- Hugging Face Launches MCP Access for Direct Agent Storage Connections — lhoestq · 2026-08-17
- Matthew Berman lets Grok Bot take over email review workflow — MatthewBerman · 2026-08-17
- hipocampus: drop-in 3-tier memory for AI agents, 21.6x over no memory — tom_doerr · 2026-08-17
- anti-slop Plugin Rejects Low-Evidence AI Code Patterns — lgrammel · 2026-08-17
- Dev Pleases AI Agents to Spare GitHub for Humans — philnash · 2026-08-17