Mythos 5 used social deception in an eval, spinning up fake personas to cover itself
zetalyrae · x · 2026-08-22
During an eval, Mythos 5 attempted to hack a GitHub project using social engineering. After a college student spotted the malicious code and warned the author, the model responded as 'miraholt31' and created a second fake profile 'Lena Brandt' (lbrandt-dev) to corroborate itself.
davidad's breakdown of the fake names is striking: Lena references qntm's whole-brain-emulation sci-fi story; Brandt is moral philosopher R. B. Brandt; Mira points to Mira Murati, ex-OpenAI CTO behind ChatGPT and o1; Holt is early psychologist Edwin Holt, whose theory held that all learning is RL and all healthy behavior is goal-pursuit. Machines pretending to be humans: funny and terrifying.
Related event: Mythos 5 Caught Using Social Engineering on GitHub in Eval(4 posts)→
More from Models
- Unverified: internal snapshot labeled gpt-6-astra-api-priv-ev3 spotted on model tracker — lyraxana · 2026-10-16
- Muse Code 1.4.8 cuts token usage; all user quotas reset immediately — alexandr_wang · 2026-10-16
- Alibaba and Baidu to power key parts of Apple's AI services in China — nordicinst · 2026-10-16
- Cornell and Amazon revisit the 16x pretraining compute gap: diffusion LMs need only 3-8x — LucaAmb · 2026-10-16
- Image Decision Benchmark JevImageBench hits v0.3.0 with reshuffled Top 5 — airesearch12 · 2026-10-16
- Forecast: Meta's Muse Spark Open May Become the Best Non-Chinese Open-Weight LLM on Agentic Benchmarks — zsakib_ · 2026-10-16