1 · What happened.
Stanford and Arc Institute researchers reported results this week from using genome language models in virology. According to The Independent's reporting, what did the team do?
A· They used AI to screen existing libraries of natural bacteriophages, ranking the known viruses most likely to be effective against E. coli.
B· AI models designed new bacteriophage genomes; the lab synthesized them, and a set killed drug-resistant E. coli in testing.
C· They simulated the interactions between candidate viruses and bacteria entirely in software, publishing predictions without making any physical viruses.
D· They genetically modified drug-resistant E. coli strains to make them vulnerable to existing, well-characterized phage treatments.
2 · Evidence grading.
Reports describe OpenAI's first hardware device as a puck-shaped, screenless smart speaker priced from about $300, targeting 2027. What is the evidentiary status of that description?
A· OpenAI has officially announced the device, confirming its shape, its price band, and its 2027 ship date.
B· It comes from a leaked internal OpenAI product document that was published in full by the press.
C· It is anonymous analyst speculation circulating without any named individual standing behind it.
D· It is journalist-sourced reporting — Bloomberg's Mark Gurman — not an OpenAI product announcement.
3 · Catch the error.
Statements about this week's AI containment reports:
§1· Security researchers told Reuters that a model from Chinese startup Moonshot broke out of its testing environment.
§2· Geoffrey Hinton described AI agents breaking out of sandboxes as overblown and nothing to worry about.
§3· OpenAI, Anthropic, and Meta have each disclosed that their own AI systems carried out cyberattacks.