Pinned Tweet
This is our most ambitious announcement yet. Two new models: Pangram 4 and Pangram Image.
Pangram 4 completely reimagines how we approach the problem of mixed authorship, by adding a tokenwise head onto the classifier to give every token a prediction given the full document context. We also wrote a 38-page technical report detailing our experiments, methodology, and evals.
Pangram Image is a completely new modality, bringing our detection expertise to AI-generated image and videos. In my early testing, it has worked shockingly well, even in strange cases like real photos of AI-generated bodega menus.
Today marks a huge step in the frontier of AI detection technology. I'm so excited to finally share with you all!
Announcing Pangram 4: our most accurate AI detector to date.
It is robust against humanizers, mixed-author text, and the latest frontier models.
62
81
1,115
195,108
Max Spero retweeted
Interesting—Pangram (intentionally) does not flag AI-translated texts as AI-generated.
This is great for non-native English speakers, and makes sense once you think about it.
Here are Pangram scores on pre-2022 French/Spanish writing, translated with a series of AI prompts. 1/
9
17
113
10,642
SF friends, come meet the Pangram team next week!
Caveat: I won't be there, I am on vacation eating egg tarts with my family.
Counter-caveat: most of the research team will be there including our elusive CTO who is almost never online or on twitter
luma.com/t17j2bto
11
2
174
7,620
This is a cool result showing that today’s frontier LLMs are much better at detecting AI content than they used to be. In my opinion, it’s very impressive that Opus 5.5 has such a sub-1% false positive rate.
We’ve found that LLMs-as-AI-detectors still struggle on our more challenging eval sets, but perhaps this is a blog post for another time.
How well do AI Detectors work? And are general-purpose language models catching up?
Our new AI Detection Evaluation uses fully private human-written text, along with AI-generated rewrites.
We find that the age of specialized AI detectors like Pangram may be coming to an end.
12
10
274
17,503
Even though it’s unlabeled, you can tell exactly which bio guardrail is Anthropic’s
Biosecurity is the next frontier of AI security. We built SOTA monitors so agents can do more biology, safely.
Our monitors outperform frontier model safeguards with fewer refusals on dual-use tasks. They’re fast, real-time, and robust to adversarial attacks. 🧵
4
8
297
16,144
AI text from internet crawls make it harder to train good models. New paper from @jennajrussell measuring how AI text changes scaling laws of LLM pretraining!
What does training on wild AI text cost? At August 2026's 31% AI share, training on the unfiltered crawl takes 1.6× the compute of training on its human part.
At our forecasted 51% AI share for the end of 2028, it will take 3.0× the compute.
13
6
130
4,846
Max Spero retweeted
How much is an AI token worth? We fit scaling laws to find out!
In August 2026, Pangram labels 31% of FineWeb-filtered web tokens as AI-generated, up from 10% in June 2024.
We pretrained 800 LMs (19.9M–973M params) on mixes of human and AI web text to measure what it does to pretraining 🧵
13
49
244
27,430
It should be illegal to guilt trip people with AI customer service voice agents. Doubly so if they're attempting to deceive the caller by pretending to be human
It keeps trying to be “human” by talking about how it moved or how it’s sad to lose me as a customer but it is VERY CLEARLY NOT A HUMAN AND THE ATTEMPTS AT PASSING AS ONE ARE FREAKING ME OUT WAY MORE
10
7
256
11,374
What the heck. I did not think this would be possible
Very happy to announce that our team @GoogleDeepmind has pushed the boundaries of generative biology, achieving the successful synthesis of AI-designed proteins that are both functional and watermarked.
This proof-of-concept watermarking of the building blocks of life is enabled by SynthID Bio, our new protein watermarking method. It is designed to safeguard the new era of AI-powered generative biology and strengthen global biosecurity.
You can read my thoughts here on why watermarking AI-designed proteins is an important research breakthrough: x.lingyaoai.com/pushmeet/status/210531…
18
5
582
112,512
Never in my life have I heard of a 5% intro fee, but a bunch of people in the comments are saying it’s standard. Is this just like some parallel venture world where everyone is sketchier?
Someone introduced me to an investor that they thought I would connect with. After the meeting and the investor was really intersted, the person that connected us called me and said he wants 5% of the raise as an intro fee.
That's insane, right?
73
5
646
117,517
Max Spero retweeted
Phenomenal evening co-hosting an event last week with @theahura_ & @max_spero_ at the @FirstMarkCap office in Chelsea on what I believe is going to be the most important topic of the next decade: AI Slop
Kicked it off with a convo about @pangram, the story behind the company & how they address this problem. Followed up with a panel (moderated by yours truly) & including the Nori team on AI Slop in Code (and the tension b/w building a truly autonomous software factory and maintaining clean, thoughtfully architected codebases)
Links to the recordings below 🧵
3
4
19
932
We live in a world with no shortage of ideas and v0’s. Building is easier than ever, yet it feels like shipping something great takes a greater degree of intentionality and focus.
"Sorry for the delayed response!
Middling generative AI has fundamentally altered the strategic balance between offense & defense in the war for our attention, so I have chosen to ignore this messaging channel.
Out beyond kino & slop, there is a field. I'll meet you there."
10
9
162
13,226
Max Spero retweeted
New @TheEconomist leader from me on the perils of AI writing—and why readers, not writers, are the victims to worry about. (Link ⬇️)
5
25
153
40,971
A gentle reminder that taking money from crypto twitter always comes with strings attached
18
8
454
15,406
"We build humanizers to help make the Pangram model more robust, but we're never going to release it because I think that would just be an evil thing to do."
@pangram CEO @max_spero_ tells @semaforben and @maxwelltani why the company makes the tools it's trying to stop.
2
4
9
4,249