On October 13, ABC reported on an open scientific letter about cyber-capable AI. The primary text is dated October 2026, without a precise day: October 13 dates the reporting, not an established adoption date. ABC described 350 scientists at that point; that reported count should not be confused with a subsequently updated signatory list.
The letter asks organisations developing, deploying or testing these models to adopt high-assurance containment, initially favouring disconnection from the internet. It also proposes an international body to track containment failures and establish, with governments and specialists, practices that could eventually permit broader access. Incidents should be reported to that body and ultimately made public.
These are scientific recommendations, not legislation taking effect. Their argument is that model alignment and environment security must improve together: more predictable behaviour does not replace a technical boundary. For connected agents, the trade-off between permitted capabilities and possible consequences still needs assessment; the letter does not establish that a single mechanism secures every use case.