Sep 10, 2026 · 28 min
AI platforms face a safety gap and a security breach
OpenAI's New Image Model and AI Safety
The episode connects long-term fears about controlling advanced AI with immediate risks from compromised accounts and stolen access tokens.
- 1Apple’s folding iPhone Duo signals how new hardware may become a vehicle for anticipated AI features.
- 2The unresolved alignment problem leaves open whether leading AI companies have a credible plan for superintelligence risks.
- 3Stolen sessions and hijacked OAuth tokens reportedly drained Claude subscribers’ allowances, exposing weaknesses in AI platform security.
Don't miss
The episode’s sharpest turn comes when abstract superintelligence fears give way to reports of stolen sessions draining subscribers’ Claude allowances.
The brief
Apple’s newly released iPhone Duo opens the episode, with its dual-screen design framed alongside anticipated AI capabilities and the broader direction of consumer technology.
The discussion then shifts from product launches to the unresolved challenge of aligning superintelligent systems, questioning whether leading AI companies have a credible safety plan.
Comments about catastrophic outcomes make the alignment debate concrete: the central uncertainty is not whether advanced systems could be powerful, but whether their goals can remain controlled.
The final segment turns to an immediate threat, as reports describe stolen sessions and hijacked OAuth tokens draining Claude subscribers’ allowances.
Anthropic reportedly confirmed infostealer-related abuse, leaving unexplained token usage as a reminder that AI safety also depends on ordinary account security.
iPhone Duo
Apple Inc.
OpenAI
ChatGPTListen to the full episode and explore every guest, topic, and moment on PodLume.