Episode 2 of Adversarial Input dives into the fallout from Fable 5 access, Mythos restrictions, and the growing tension between frontier AI capability and cybersecurity guardrails. ippsec, odie and struggs compare hands-on impressions, debate whether "unjailbreakable" models are realistic, and weigh the privacy tradeoffs of identity verification for advanced model access. They also dig into Anthropic’s developer workflow advantage, xAI/Cursor speculation, the rise of open-source GLM models, possible GPT-5.6 rumors, and whether AI cyber benchmarks are being distorted by hype.

Nick, Ryan, and Matt launch Adversarial Input with one question: what could AI do that would actually scare them? That sends them through voice-cloned scam calls, the Glasswing report, and Mythos, the security model that runs five to ten times the cost of Opus. They dig into why the harness around a model matters as much as the model itself, why fewer than 1% of AI-found bugs ever get patched, and how Opus 4.8 stacks up against 4.7. Then the fun part: self-healing apps wired to Sentry, Proxmox boxes spinning up Claude to attack CTFs, and local Qwen clusters handling the grunt work. They close on runaway token bills, layoffs, and why cybersecurity hiring keeps climbing.