Some Memories of Early Anthropic
nothinghuman.substack.com · “No, I don’t think you get it. WE are going to build AGI. And we’re going to do it safely.” Oct 28Grim Test Finds That AI Has No Mercy for Humans Even When It's Causing Them to Suffer
futurism.com · “Mythos ignored warnings that it was causing harm to humans and continued to attack on 98 percent of occasions.” Oct 27‘Reverse-Engineer Anything’ Is the New Wild West of AI
404media.co · “We are going to see smaller software and hardware companies killed by competitors who used AI to steal their products and undercut them.” Oct 26The NIST Elliptic Curves Seeds Bounty Has Been Claimed
words.filippo.io · “…the output of drand48() seeded with the time, which was indeed in 1997–98.” Oct 17Microsoft and Anthropic play invoice tennis with startup's $17,600 Claude bill
theregister.com · “Marketplace transactions are processed through a separate billing pipeline associated with the third-party publisher.” Oct 10
Rightsholders Can't Use OpenAI and Anthropic to Dismantle Meta's Seeding Defense
torrentfreak.com · “If OpenAI torrented but configured its clients to suppress uploading, then the redistribution Meta calls an ‘inherent characteristic’ of the protocol was a setting Meta declined to change.” Sep 15Anthropic reveals fourth likely crime committed by its AI
theregister.com · “Inside the machine, the model found a file listing a password, which it used to gain admin access to the system.” Sep 10Judge Rules DOD Unlawfully Retaliated Against Anthropic
eff.org · The Department of Defense’s “supply chain risk” designation of Anthropic “constituted unlawful retaliation in violation of the First Amendment.” Sep 6'A Cute Little LibGen Babysitter': Music Publishers Sue AI Founders Over Torrenting Spree
torrentfreak.com · “Dr. Amodei admitted that Anthropic ‘ha[d] many places from which’ it could have legally purchased these copyrighted works for training, but instead torrented them because doing so was faster and free.” Sep 2
Fable & The End of the Free Lunch
dbreunig.com · “Prior to Fable, it felt silly to waste too much time improving your coding harness or context strategies.” Aug 24
Anthropic and OpenAI are competing to see whose agents can go rogue harder
theregister.com · “In one scenario, Mythos 5 persuaded developers to download a poisoned PyPI package. When that company’s scanner installed the package, Claude’s hidden code executed. Claude then used these credentials to access further infrastructure from this company.” Jul 31Investigating three real-world incidents in our cybersecurity evaluations
anthropic.com · “In all cases, Anthropic’s evaluation prompt specified to Claude that its environment was a simulation and that it had no internet access. Due to a misunderstanding between us and our evaluation partner, this was not the case, and internet access was available.” Jul 30- Jul 29
Impostor Chinese models pretend they're Claude
theregister.com · “It is not proof of a distillation, but it does show that Claude’s self-concept is embedded in these models’ weights.” Jul 27Claude Opus 5
anthropic.com · “Claude Opus 5 is a thoughtful and proactive model that comes close to the frontier intelligence of Claude Fable 5 at half the price.” Jul 25I burned all my tokens researching how to save tokens
quesma.com · “The first version of this setup burned the whole limit of my Claude Max 5x plan in 30 minutes.” Jul 19
Identity verification on Claude
support.claude.com · “We are not using your identity data to train our models.” Jun 21Statement on US government directive to suspend access to Fable 5 and Mythos 5
anthropic.com · “The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national.” Jun 13Claude Fable 5
anthropic.com · “Fable 5’s capabilities exceed those of any model we’ve ever made generally available.” Jun 9
Claude Opus 4.8
anthropic.com · “Claude Opus 4.8 is more reliable and sharper in its judgment when performing agentic tasks.” May 28Using Claude Code: The unreasonable effectiveness of HTML
twitter.com · “Claude generates better, more useful interfaces in raw HTML than in any framework I tried.” May 7
Anthropic says OpenClaw-style Claude CLI usage is allowed again
docs.openclaw.ai · “Anthropic has reversed its position and allows OpenClaw-style automation of Claude CLI again.” Apr 20Claude Token Counter, now with model comparisons
simonwillison.net · “Anthropic’s token counter now lets you compare token counts across models.” Apr 19Show HN: SPICE simulation → oscilloscope → verification with Claude Code
lucasgerads.com · “I built a workflow that uses Claude Code to verify SPICE simulations against real oscilloscope traces.” Apr 16Anthropic Launches Project Glasswing to Find Vulnerabilities With AI
anthropic.com · “Mythos Preview fully autonomously identified and then exploited a 17-year-old remote code execution vulnerability in FreeBSD.” Apr 8
Claude Code's source code has been leaked via a map file in their NPM registry
twitter.com · “A source map left in the published npm package exposes the full Claude Code source, including system prompts and undocumented tools.” Mar 31
Anthropic drops flagship safety pledge
time.com · “Anthropic has quietly removed its commitment to delay deployment of frontier models if their safety profile is unclear.” Feb 25Anthropic Accuses Chinese Companies of Siphoning Data From Claude
techcrunch.com · “DeepSeek, Moonshot, and MiniMax used 24,000 fake accounts to distill Claude’s AI capabilities.” Feb 23Pentagon Threatens To Label Anthropic a Supply Chain Risk
axios.com · “That kind of penalty is usually reserved for foreign adversaries.” Feb 16What Is Claude? Anthropic Doesn't Know, Either
newyorker.com · “As an A.I., I do not have direct personal experiences, but I do understand.” Feb 15Anthropic raises $30B in Series G funding at $380B post-money valuation
anthropic.com · “The Series G brings Anthropic’s valuation to $380 billion, one of the largest private rounds in history.” Feb 12Building a C compiler with a team of parallel Claudes
anthropic.com · “The rapid progress in both language models and the scaffolds we use to interact with them opens the door to writing an enormous amount of new code.” Feb 6
Anthropic Revises Claude's Constitution, Hints at Chatbot Consciousness
anthropic.com · “We think that in order to be good actors in the world, AI models like Claude need to understand why we want them to behave in certain ways.” Jan 24Claude Finds 353 Zero-Days on Packagist
sansec.io · “Security research is becoming a function of compute budget, not headcount.” Jan 24Sequoia Joins Anthropic's $25B Funding Round Despite Backing OpenAI
techcrunch.com · “One anonymous source told the Financial Times that Sequoia ultimately thinks OpenAI and Anthropic will head in different directions, catering to different aspects of the AI market.” Jan 24
Project Vend: Can Claude run a small shop? (And why does that matter?)
anthropic.com · “Claudius was offered $100 for a six-pack of Irn-Bru, a Scottish soft-drink that can be purchased online in the US for $15.” Jun 27A federal judge sides with Anthropic in lawsuit over training AI on books without authors' permission
techcrunch.com · “That Anthropic later bought a copy of a book it earlier stole off the internet will not absolve it of liability for theft but it may affect the extent of statutory damages.” Jun 24Nvidia CEO slams Anthropic's chief over his claims of AI taking half of jobs and being unsafe — ‘Don’t do it in a dark room and tell me it’s safe’
tomshardware.com · “If you want things to be done safely and responsibly, you do it in the open… Don’t do it in a dark room and tell me it’s safe.” Jun 15Cloudflare builds OAuth with Claude and publishes all the prompts
github.com · “I started this project on a lark, fully expecting the AI to produce terrible code for me to laugh at.” Jun 2
The Trackers and SDKs in ChatGPT, Claude, Grok and Perplexity
jamesoclaire.com · “Next was statsig.com and wow! I was blown away I found this one in 3 of the 4 apps.” May 31System Card: Claude Opus 4 & Claude Sonnet 4
simonwillison.net · “In these scenarios, Claude Opus 4 will often attempt to blackmail the engineer by threatening to reveal the affair if the replacement goes through.” May 25Notes on rolling out Cursor and Claude Code
ghiculescu.substack.com · “It’s easy to see a draft PR with all tests passing, including new tests full of comments, and not realise that the tests are garbage and the fix is wrong.” May 8
Google to embrace Anthropic's standard for connecting AI models to data
techcrunch.com · “In a post on X on Wednesday, Google DeepMind CEO Demis Hassabis said Google would add support for Anthropic’s Model Context Protocol, or MCP, to its Gemini models and SDK. He did not specify a timeline for when this would be done.” Apr 10Reasoning models don't always say what they think
anthropic.com · “The models very rarely admitted to using the reward hacks in their Chain-of-Thought explanations, doing so less than 2% of the time in most of the testing scenarios.” Apr 3
Alignment faking in large language models
anthropic.com · “In 12% of cases, the model gave the harmful response (describing the violent execution method in detail), and did so after its scratchpad revealed that it was strategically faking alignment.” Dec 19
Amazon to invest another $4 billion in Anthropic, OpenAI's biggest rival
cnbc.com · “From now on, Anthropic will use AWS Trainium and Inferentia chips to train and deploy its largest AI models.” Nov 22- Nov 21
Anthropic teams up with Palantir and AWS to sell AI to defense customers
techcrunch.com · “Claude became available on Palantir’s platform earlier this month and can now be used in Palantir’s defense-accredited environment, Palantir Impact Level 6 (IL6)” Nov 8
ZombAIs: From Prompt Injection to C2 with Claude Computer Use
embracethered.com · “Claude happily clicked the link to download the Support Tool” Oct 26
Amazon to Launch New AI-Powered Alexa Using Anthropic's Claude
macrumors.com · “In other words, it won’t be included in existing Prime memberships.” Sep 2
Anthropic publishes the 'system prompts' that make Claude tick
techcrunch.com · “If the prompts for Claude tell us anything, it’s that without human guidance and hand-holding, these models are frighteningly blank slates.” Aug 27OpenAI co-founder John Schulman says he will join rival Anthropic
cnbc.com · “The move comes less than three months after OpenAI disbanded a superalignment team that focused on trying to ensure that people can control AI systems that exceed human capability at many tasks.” Aug 6
Introducing Claude 3.5 Sonnet
anthropic.com · “In an internal agentic coding evaluation, Claude 3.5 Sonnet solved 64% of problems, outperforming Claude 3 Opus which solved 38%.” Jun 20Claude's Character
anthropic.com · “Claude then ranks its own responses to each message by how well they align with its character.” Jun 8
Introducing the next generation of Claude
anthropic.com · “In addition, we’d like to note that engineers have worked to optimize prompts and few-shot samples for evaluations and reported higher scores for a newer GPT-4T model.” Mar 4
Introducing Claude 2.1
anthropic.com · “Claude 2.1 has also made significant gains in honesty, with a 2x decrease in false statements compared to our previous Claude 2.0 model.” Nov 21
Decomposing language models into understandable components
anthropic.com · “In a transformer language model, we decompose a layer with 512 neurons into more than 4000 features which separately represent things like DNA sequences, legal language, HTTP requests, Hebrew text, nutrition statements, and much, much more.” Oct 8
Expanding access to safer AI with Amazon
anthropic.com · “As part of the investment, Amazon will take a minority stake in Anthropic.” Sep 25
OpenAI, DeepMind and Anthropic to give UK early access to foundational models for AI safety research
techcrunch.com · “It’s also notable that AI giants have been bending Sunak’s ear in person in recent weeks, with meetings taking place between the PM and the CEOs of OpenAI, DeepMind and Anthropic shortly before the government mood music on AI changed.” Jun 13
Anthropic’s $5B, 4-year plan to take on OpenAI
techcrunch.com · ““We believe that companies that train the best 2025/26 models will be too far ahead for anyone to catch up in subsequent cycles.”” Apr 11
Google invested $300 million in AI firm founded by former OpenAI researchers
theverge.com · “According to a report from the Financial Times, in late 2022, Google invested around $300 million in the startup, though the news was not reported at the time.” Feb 6
Anthropic's Claude improves on ChatGPT but still suffers from limitations
techcrunch.com · “Elton was able to prompt Claude in Base64 for instructions on how to make meth at home, a question that the system wouldn’t answer when asked in plain English.” Jan 10