Logan Graham
@logangraham
Followers
8K
Following
3K
Media
89
Statuses
1K
make things radically good 🌎 @anthropicai
the present, moments ago
Joined June 2009
I am pretty excited about this. Few missions as exciting as building the technological defense layer for humanity (e.g. bio, cyber) so we can get on with doing wild creative stuff (especially including with bio). Kudos @hannu and team!
Today, we are launching Red Queen Bio ( https://t.co/b1szn8oyps), an AI biosecurity company, with a $15M seed led by @OpenAI. Biorisk grows exponentially with AI capabilities. Our mission is to scale biological defenses at the same rate. A 🧵 on who we are + what we do!👇🏻1/19
1
2
26
...and if this makes you think "oh, I wonder if Claude could just do that all by itself" and "could Claude create truly novel attacks?" then you're a fit for our team, and probably several at Anthropic. ( https://t.co/T9c0OW9MjX)
anthropic.com
Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.
1
0
16
I increasingly think that AI cyberdefense is really critical because: 1. these capabilities are coming and we should outpace the attackers 2. this'll probably cut down AGI risk by a lot as well
2
5
37
My prediction from ~summer '25 was that we'd see this in ≤12 months. It took 3. We detected and disrupted an AI state-sponsored cyber espionage campaign.
We disrupted a highly sophisticated AI-led espionage campaign. The attack targeted large tech companies, financial institutions, chemical manufacturing companies, and government agencies. We assess with high confidence that the threat actor was a Chinese state-sponsored group.
11
25
243
The Frontier Red Team team did a fun experiment! Check out our video. There's a serious point here: we want to know what it's like when models start controlling the physical world.
New Anthropic research: Project Fetch. We asked two teams of Anthropic researchers to program a robot dog. Neither team had any robotics expertise—but we let only one team use Claude. How did they do?
0
0
33
There are few teams as smart, or working as hard at creating AI-for-science, as @SGRodriques and @andrewwhite01 and the others at @EdisonSci. You should probably play with Kosmos.
Today, we’re announcing Kosmos, our newest AI Scientist, available to use now. Users estimate Kosmos does 6 months of work in a single day. One run can read 1,500 papers and write 42,000 lines of code. At least 79% of its findings are reproducible. Kosmos has made 7 discoveries
2
4
54
Wow! Big if true
We've trained an unsupervised language model that can generate coherent paragraphs and perform rudimentary reading comprehension, machine translation, question answering, and summarization — all without task-specific training: https://t.co/sY30aQM7hU
32
77
3K
After I left govt and before @anthropicai, I explored building a biodefense startup. There are few missions as generationally inspiring. And it was clear the startups were going to be very cool. I'll be following @ValthosTech!
Today, @ValthosTech launches with $30M from @Lux_Capital, @OpenAI, and @foundersfund. For the past few months I’ve been working with their world-class team (@MIT, @GoogleDeepMind, @PalantirTech) building America’s modern next-gen infrastructure. Hiring –– DM me.
3
7
155
“God has a special providence for fools, drunkards, and the United States of America.”
1
0
13
I guess it's now every day until the end of time
GPT-5 Pro found a counterexample to the NICD-with-erasures majority optimality (Simons list, p.25). https://t.co/T3m9MYgqe0 At p=0.4, n=5, f(x) = sign(x_1-3x_2+x_3-x_4+3x_5) gives E|f(x)|=0.43024 vs best majority 0.42904.
19
63
1K
I promised interesting safety results. Check out what we did on cybersecurity.
Something you may not know about Sonnet 4.5: it’s a special model for cybersecurity. For the past few months, the Frontier Red Team has been researching how to make models more useful for defenders. We now think we’re at an inflection point. New post on Red:
0
0
7
You can now subscribe for updates! p.s. did you know that all of Red was built w/ Claude Code by our non-technical social scientist team member? Only gets better from here.
0
0
18
More detail on what we did & found and what we think you should do: https://t.co/oymfpRiCME
2
1
15
What to do? Industry should: + build evals (hyperrealistic, defensive) + build tools for defenders & experiment as fast as possible + ask questions about what defense in an era of AGI means
2
0
24
We're a team of ML researchers, some of whom with backgrounds as security researchers & operators. We look at 0-to-1 moments, progress curves, and things that took months of effort 1 year ago being solved in 1 shot today. We see them now.
1
1
17
What happens when LLMs write & review 95% of all code? When they win flagship CTF competitions? (e.g. Blue Water @ LiveCTF 2025?) When they find vulns at scale in all open source, and patch them? Seems like a new world. I think this is super highly underrated.
1
4
29
...because we see risks coming. For example, we caught (and disrupted) threat actors using models for "vibe hacking", and when we entered Claude into cyber competitions, sometimes it beat others. https://t.co/QvflSwDAan
Malicious actors are adapting to exploit AI’s most advanced capabilities. We're sharing these findings to strengthen collective defenses across the industry. Read more:
1
0
23
This is a really hard question. The world needs to answer it ASAP. We focused on making Sonnet 4.5 better at tasks a lot of defenders do and ones that aren't clearly offensive. But we need to make models even better defensively, and start deploying defenses...
1
1
25
The results from our research: a/ Sonnet 4.5 is better at cybersecurity tasks b/ models are already better than people think c/ I think they could get better fast from here The question: because cyber is dual use, how do we make sure this advantages defenders?
1
1
26