ℏεsam

"sandboxing is an antiquated technology to apply to agents"

That's Ali Golshan, who leads NVIDIA's open source agent runtime called OpenShell, which is now built into Anthropic's Claude Managed Agents.

I sat down with Ali at GTC Berlin to talk about where AI security is headed and what makes OpenShell different:
> agents start with zero privileges
> more access is given only if necessary
> secrets stay out of the containment, without the agent being able to access them directly

what we discussed:
1:15 Are AI security risks overblown?
6:06 Do we need new AI regulations?
7:47 What is OpenShell? And what’s the difference with a sandbox.
10:20 Would OpenShell have stopped the Hugging Face incident?
13:44 Alignment vs. Security
16:04 Advice for getting into AI safety

Huge thanks to @nvidia for inviting me and setting up this interview.

it looks like scientific discovery is turning into the real leaderboard for frontier labs.

the public doesn’t care about benchmarks, and the scores don’t really match user experience.

OpenAI is going hard on math.
Anthropic on biology.
now DeepMind and Meta too.

much better PR and also a much more exciting race.

WOW… DeepSeek Harness grew 545% in daily active users in just 36 days, now reaching 1.54 MILLION.

its bet on plugins and a highly modular harness design seems to be paying off:

“Since launch, the number of third-party plugins has grown to roughly 35,000, and about 60% of active users have built or installed at least one.”

well done.

funny thing is, Anthropic actually wanted to IPO in October, until some of the advisors said waiting until November would be better to show “a strong competitive position”

now the IPO valuation expectations have been slashed from $2T to as low as $1.1T

what an expensive advice

> “what happened to the bitter lesson?”

people are genuinely trying to explain the bitter lesson to the guy who wrote it.

it’s called the bitter lesson
not the dumb lesson.

the bitter lesson says general methods that scale with compute win in the long run. you still have to invent those methods and improve the existing ones.