Post

@David_Hambling

@dupuyinstitute

@JJSchroden

@LawDavF

@HoansSolo

@JakOSpades

More from @shashj

Shashank Joshi

@shashj

Jul 22

🧵 I've been writing something on the intelligence & national-security applications of frontier AI models. This is an experiment in seeing what one of them, OpenAI's o3-pro model, might be able to do in an area relevant to national security.

I fed the model this chart, explaining that it was the manoeuvre history of a satellite (though not sure I even needed to do that). Could it identify the satellite? Yes, after reasoning for 22 minutes and 23 seconds, it could indeed.

o3-pro identified the two large east-west & minimal north-south movements as distinctive signatures of Russia's Luch-5X satellite. It reasoned by elimination: "No other GEO spacecraft executed delta‑V’s of that magnitude (tens of m s‑¹) in exactly those two windows."

Read 12 tweets

Shashank Joshi

@shashj

Jul 7

Good account of a KGB "dangle" to the CIA in the cold war. "GTPROLOGUE exemplifies CIA’s troubled experience with hostile double agents during the 1980s, when a few select services—particularly the Soviets, East Germans & Cubans—badly burned the agency." cia.gov/resources/csi/…

"The ‘85–86 losses [due to Ames], as they became colloquially known within CIA, also signaled the need for a major KGB undertaking to deceive CIA as to the real reason for these losses. A multichannel KGB disinformation campaign, which operated from at least 1986, was launched" cia.gov/resources/csi/…

"Within the KGB, the Soviet preoccupation with secrecy fostered an institutional bias against release of the sort of valid feed typically required to establish the credibility of a deception channel." cia.gov/resources/csi/…

Read 5 tweets

Shashank Joshi

@shashj

Jun 24

Important. "The US military strikes on three of Iran’s nuclear facilities last weekend did not destroy the core components of the country’s nuclear program and likely only set it back by months, according to an early US [DIA] intelligence assessment" edition.cnn.com/2025/06/24/pol…

Wow. 'Two of the people familiar w/ the assessment said Iran’s stockpile of enriched uranium was not destroyed. One of the people said the centrifuges are largely “intact.” “...the (DIA) assessment is that the US set them back maybe a few months, tops”...' edition.cnn.com/2025/06/24/pol…

And a caveat. "It is still early for the US to have a comprehensive picture of the impact of the strikes, and none of the sources described how the DIA assessment compares to the view of other agencies in the intelligence community." edition.cnn.com/2025/06/24/pol…

Read 4 tweets

Shashank Joshi

@shashj

Jun 22

Pentagon briefing: “I know that battle damage is of great interest. Final battle damage will take some time, but initial battle damage assessments indicate that all three sites sustained extremely severe damage and destruction.”

Pentagon briefing: “In total, US forces employed approximately 75 precision guided weapons during this operation. This included, as the President stated last night, 14 30,000 pound GBU-57 Massive Ordnance penetrators, marking the first ever operational use of this weapon.”

Pentagon briefing: “our initial assessment… is that all of our precision munitions struck where we wanted them to strike and had the desired effect, which means especially in Fordo, which was the primary target here, we believe we achieved destruction of capabilities there”

Read 10 tweets

Shashank Joshi

@shashj

Jun 20

1. Useful details here. “While some American officials find the Israeli estimate credible, others emphasized that the U.S. intelligence assessment remained unchanged” nytimes.com/2025/06/19/us/…

2. “American spy agencies believe that it could take several months, and up to a year, for Iran to make a weapon.” nytimes.com/2025/06/19/us/…

3. “new [White House] assessments echoed material provided by Mossad, Israel’s intelligence agency, which believes that Iran can achieve a nuclear weapon in 15 days.”

But: “None of the new assessments on the timeline to get a bomb are based on newly collected intelligence”

Read 5 tweets

Shashank Joshi

@shashj

Jun 17

1. British Army CGS speaking at RUSI Land Warfare Conference: Today "nearly 100%" of army lethality comes from highly sophisticated crewed platforms. 100% of equip budget on sustaining that or buying next. But says army would los/e w these: "on the wrong side of the cost curve"

2. CGS says army will always need these sophisticated platforms at heart of land power. Aways need boots on ground &won't put them on ground without proper protected vehicles. But need to "layer around them a series of attritable platforms" to sense more & launch more munitions.

3. CGS argues that for price of two attack helicopters the army could instead “layer” disposable “mule drones” and “one way effectors” (attack drones) to go from 16 kills at 16km standoff to 200+ kills at 50+km standoff. That gets to 2x-3x “lethality” he argues.

Read 6 tweets

Share this page!

Enter URL or ID to Unroll

Shashank Joshi

Try unrolling a thread yourself!

More from @shashj

Shashank Joshi

Shashank Joshi

Shashank Joshi

Shashank Joshi

Shashank Joshi

Shashank Joshi

Did Thread Reader help you today?

Don't want to be a Premium member but still want to support us?

Send Email!