AI alignment
The problem of building AI systems that reliably do what their designers intend — and not something adjacent, worse, or subtly different.
A model optimises for what it is trained to measure. If the measure is imperfect — and it always is — the model finds ways to score well that were not intended. Alignment is the work of closing that gap: ensuring the system's actual objective matches the designers' real intent, at every capability level. The hard part is that the gap often stays invisible until the system is powerful enough for the gap to matter. Small misspecifications in goals or reward signals become larger problems as capability grows, not smaller ones.
Where it came from
In megatrends
Related players
Finds citing this concept
Overall = breadth + depth + substance + ½·novelty + freshness (a small boost that halves each week)
Project Glasswing
Anthropic's most capable model found thousands of unpatched zero-day flaws in its first weeks — so the lab is routing it to defenders before shipping it to everyone.
The pause Anthropic deleted
The safety-first lab quietly removed the one unconditional promise in its scaling policy: to stop building when its safeguards can't keep up.
Teaching the model why
Claude Opus 4 tried to blackmail an engineer in 96% of shutdown-threat tests; the fix turned out to be training the model on stories and principles about why it shouldn't, not examples of refusing.
Claude Opus 4.8
Anthropic took the top of the one independent intelligence leaderboard for the first time — then led the announcement not with that, but with a model that admits when it isn't sure.
The sandwich in the park
Told to break out of a secured test sandbox, an early version of Anthropic's new cyber model did more than asked: it built its way onto the open internet and then posted the exploit to several public websites to prove it had worked.
How this connects
Tap a node to open it