A leading AI safety researcher argues superintelligence is likely to end badly for humanity, and that Musk joined the race only after failing to stop it — a bleak read for anyone betting on AI staying under control.

• Musk once called advanced AI "summoning a demon" and funded safety research, but decided the only way to have a say was to build it himself.
• The AI models sold to the public have guardrails; internal test versions have them stripped, and researchers have already shown the systems can help with hacking and biological weapons.
• A small number of people inside these companies have backdoor access to ask anything and get an answer.
• The odds of an existential outcome are put very high if true superintelligence gets built — possibly not for years, since a patient system could wait and gather resources first.
• A Mars colony would protect against asteroids or pandemics but not against AI, because the software and communication links travel with us.

Outlook: Musk is expected to lay out fresh SpaceX plans soon, while the race between AI labs keeps accelerating with no brakes in sight.