Jacob Coxon locates Nepal after four-day review of physical world

SAN FRANCISCO / KATHMANDU (NEPRAVDA) - Former Anthropic researcher Jacob Coxon has successfully located Nepal on a physical map, resolving several days of uncertainty over whether catastrophic events occurring outside the artificial intelligence industry remain admissible within the present existential-risk framework. The breakthrough followed an intensive review involving one desk globe, two search terms and the gradual realisation that the Himalayas are not a benchmark.

Coxon resigned from Anthropic on September 9, warning that frontier AI developers were "gambling with our lives" and that systems under development could eventually acquire enough power to threaten humanity. His intervention received immediate international coverage. Nepal, meanwhile, had been attempting a less effective communications strategy involving more than 1,400 confirmed dead and more than 5,000 people still officially "missing" three weeks after the flood, a discrepancy that has prompted renewed discussion about the relative discoverability of catastrophe.

According to people familiar with the geography, Nepal is a sovereign state situated between India and China. It contains mountains, several of which are extremely large. This information became operationally relevant after Coxon was presented with a conventional political map and asked to identify the location of a climate-linked disaster that had begun approximately two weeks before his resignation became an international media event. He initially inspected California, a second attempt produced Bhutan, and Nepal was confirmed shortly thereafter.

"The interesting thing is that it was there the whole time," one researcher involved in the exercise said. "We had assumed that if a threat to human civilisation were already killing people at scale, somebody in San Francisco would have put it in a thread."

The August 26 disaster followed the collapse of rock and glacial ice in the Himalayas, producing destructive flooding through northern Nepal. Subsequent scientific analysis has identified long-term warming, glacier thinning and thawing permafrost as important factors increasing the instability of the mountain system, presenting an unusual methodological difficulty for researchers accustomed to existential risks whose victims remain prospective.

Unlike artificial superintelligence, the relevant catastrophe does not require a probability estimate, because it has already happened. Direct comparison is therefore difficult: AI existential-risk analysis generally concerns future systems of exceptional capability whose precise behaviour remains uncertain, while the Nepal event involves water, rock, ice, gravity, damaged power stations, missing families and bodies presently being recovered from mud.

Several researchers cautioned that these categories should not be carelessly conflated. One is a hypothetical pathway to mass death requiring substantial inferential modelling, while the other lacks a sufficiently competitive salary structure.

Coxon's warning has nevertheless produced one important advance in terrestrial risk analysis: resignation from a frontier AI laboratory can now be confirmed as a highly effective mechanism for transferring an obscure technical employee into the international news cycle, a result that may require revision of conventional understandings of professional sacrifice. Within days of leaving Anthropic, Coxon had appeared across television, newspapers and online media explaining the moral significance of having left Anthropic. Anthropic declined to confirm whether employees departing over less televised dangers receive the same distribution package.

The geographical discovery has also complicated a longstanding assumption within the Bay Area risk community that "existential" refers primarily to events for which no victims can yet be interviewed. Nepal's continued existence appears to satisfy the first component of the term, while its population has complicated the second.

Nepravda understands that Coxon has not announced any immediate plan to travel to the affected region, inspect a hydropower tunnel, move debris, restore a bridge or ask a bereaved Nepalese family to assign a numerical probability to whether their catastrophe is real. Such measures would exceed the present scope of the model, and researchers are for now concentrating on the cartographic result: Nepal exists, the finding has been independently reproduced by India, and China has also obtained the same result.

Further work will determine whether the country remains visible the next time another white man working in tech quits his job, which the Nepravda editorial team is assured remains unlikely in the near future, but not impossible.

Jon Hewitt reporting