Shakeel

Editor + founder @ReadTransformer, covering AI policy. Prev: @TheEconomist, @Protocol, @finimize

[Shakeel @ShakeelHashim]

As I said on @LBC: AI companies can no longer reliably control their systems.

[Nat Purser @NatPurser]

every day they wake up and write 100 new auditing bills and then i clock into my shift at the auditing bill analysis factory

[Matthew Khoriaty @ AI Safety Offices in Berkeley @KhoriatyMatthew]

6-word horror story I experienced at the office today:
"Ajeya Cotra seminar cancelled, work emergency"

[Shakeel @ShakeelHashim]

Having all this in one role seems like a mistake to me

[Jessica Elgot @jessicaelgot]

Wanted: new No 10 adviser to help citizens deal with rise of AI

https://t.co/MRNLQrrCjW

[“paula” @paularambles]

so i've been putting together a list of things i’d need at home to survive a few days without power, internet, or basic services. you know, in case of earthquakes, digital or otherwise.

claude’s plan includes a hamster. i don’t have a hamster. who is he. where is he sheltering

[Shakeel @ShakeelHashim]

i want this so bad

[Tyler Nishida @TylerNishida]

im surprised that no one talking about duet's to-do list! maybe its personal preference, but its the reason i keep reaching for it.

for suggested tasks — this is much better than receiving a growth hacky chat message.

to create tasks — you can simply brain dump them!

[Joe Light @joelight]

One fun way for journalists to use AI: Ask the strongest model to which you have access to do a deep dive on your work and find where it was cited or otherwise had an impact. You'll find things you had absolutely no idea about

[Ben Smith @semaforben]

🌹🌹DSA civil war over AI, @jdcapelouto reports https://t.co/QICl6DrFwb

[Shakeel @ShakeelHashim]

This is, obviously, extremely bad. It is also quite ironic given how nakedly Politico’s owner has tried to suck up to this admin — and completely failed.

[Shashank Joshi @shashj]

Quite a letter from White House to Politico.

[Shakeel @ShakeelHashim]

@RothRottweiler Just uploaded more screenshots

[Shakeel @ShakeelHashim]

News! OpenAI's written to Gov. Gavin Newsom recommending a bunch of amendments to strengthen SB 53. Includes a bunch of stuff on AI R&D and shutdown controls.

[Shakeel @ShakeelHashim]

My usual file uploading services aren’t working, here are screenshots of the full letter

[Peter Wildeford🇺🇸🚀 @peterwildeford]

New US government guidance from CAISSI + NSA + CISA:

The US government defines an "SI control incident" where an agent takes significant uninstructed actions that break operator policies, controls, or oversight. Near-misses are in scope too.

Companies are then asked to:
- have technical controls, including rapid kill and containment mechanisms and red-teaming.

- Bring in independent external auditors to verify that control and monitoring systems work.

- Write a structured risk assessment before training runs, evals, or deployments, and proceed only if senior leadership is satisfied with the risk, informed by external evaluators where appropriate.

- Monitor every layer with both programmatic and SI-based monitors, covering trajectories, reasoning, and outputs.

- Make sure agents can't touch the monitors or logs

- Use monitor models from different model families.

- In the event of an incident, preserve weights, intermediate checkpoints, environments, and rejected tool calls.

- Keep tamper-resistant logs with reliable timestamps, including reasoning traces.

- If you use models to help investigate, cross-check with humans and with models outside the incident model's family.

- Shutdown policies calibrated to SI model capability and incident severity.

- Depending on severity, pause training, evals, or deployment of similar agents until controls are re-verified.

- When restarting, validate every containment layer and replay the incident trajectory to prove the new controls would catch it.

- Document the decision to resume and the residual risk accepted.

- Write an after-action report.

- Notify anyone compromised, including intermediate victims, and share timelines and transcripts.

- Report cyber incidents and near-misses to CISA. Share SI-specific incident and near-miss information with CAISSI and NSA SISC.

- Give third parties enough access and documentation for independent review.

[Kevin Roose @kevinroose]

Our new show with NPR, Machine Gods, is finally here! Episode 1 is live wherever you get your podcasts.

The official podcast of the singularity starts today!

https://t.co/AzdJKNuWnx

[Josh Schwerin @JoshSchwerin]

Big new additions to the @csaiporg team

[Shakeel @ShakeelHashim]

something extremely funny about using Google AI Overviews to prove her point

[Shakeel @ShakeelHashim]

"data centers cause light pollution" is a new one

[Shakeel @ShakeelHashim]

@S_OhEigeartaigh @frances__lorenz i couldn't say it better -- i'm very sorry and sending lots of love

[Shakeel @ShakeelHashim]

@coryweinberg @nytimes @theinformation huge congrats!

[Cory Weinberg @coryweinberg]

Job news: I’m joining @nytimes to cover AI & Anthropic, based again in SF. I have a feeling the job won’t be slow. It's a big responsibility.

I'm grateful for a decade @theinformation, which has given me the rare opportunity to do original journalism surrounded by good people.

[Shakeel @ShakeelHashim]

A great piece from Chris — Claude's safeguards are not very hard to circumvent. And it even writes its nefarious scheming in Claudese

[Chris Stokel-Walker @stokel]

My latest for @ReadTransformer looks at how Claude will tell you how to manipulate an election with not all that much jailbreaking https://t.co/wyFrs6RVqw