Share this page!

Enter URL or ID to Unroll

You can paste full URL like: https://x.com/threadreaderapp/status/1644127596119195649
or just the ID like: 1644127596119195649

How to get URL link on X (Twitter) App

  1. On the Twitter thread, click on or icon on the bottom
  2. Click again on or Share Via icon
  3. Click on Copy Link to Tweet
  4. Paste it above and click "Unroll Thread"!
  5. More info at Twitter Help
Anthropic Profile picture
Anthropic
@AnthropicAI
We're an AI safety and research company that builds reliable, interpretable, and steerable AI systems. Talk to our AI assistant @claudeai on https://t.co/FhDI3KQh0n.
14 subscribers
Oct 29 • 12 tweets • 4 min read
New Anthropic research: Signs of introspection in LLMs.

Can language models recognize their own internal thoughts? Or do they just make up plausible answers when asked about them? We found evidence for genuine—though limited—introspective capabilities in Claude. An example in which Claude Opus 4.1 detects a concept being injected into its activations. We developed a method to distinguish true introspection from made-up answers: inject known concepts into a model's “brain,” then see how these injections affect the model’s self-reported internal states.

Read the post: anthropic.com/research/intro…
Oct 6 • 5 tweets • 2 min read
Last week we released Claude Sonnet 4.5. As part of our alignment testing, we used a new tool to run automated audits for behaviors like sycophancy and deception.

Now we’re open-sourcing the tool to run those audits. Researchers give Petri a list of seed instructions targeting scenarios and behaviors they want to test. Petri then operates on each seed instruction in parallel. For each seed instruction, an auditor agent makes a plan and interacts with the target model in a tool use loop. At the end, a separate judge model scores each of the resulting transcripts across multiple fixed dimensions so researchers can quickly search and filter for the most interesting transcripts. It’s called Petri: Parallel Exploration Tool for Risky Interactions. It uses automated agents to audit models across diverse scenarios.

Describe a scenario, and Petri handles the environment simulation, conversations, and analyses in minutes.

Read more: anthropic.com/research/petri…
Aug 1 • 11 tweets • 4 min read
New Anthropic research: Persona vectors.

Language models sometimes go haywire and slip into weird and unsettling personas. Why? In a new paper, we find “persona vectors"—neural activity patterns controlling traits like evil, sycophancy, or hallucination. Our automated pipeline takes as input a personality trait (e.g. “evil”) along with a natural-language description, and identifies a “persona vector”: a pattern of activity inside the model’s neural network that controls that trait. Persona vectors can be used for various applications, including preventing unwanted personality traits from emerging. We find that we can use persona vectors to monitor and control a model's character.

Read the post: anthropic.com/research/perso…
Jul 29 • 8 tweets • 3 min read
We’re running another round of the Anthropic Fellows program.

If you're an engineer or researcher with a strong coding or technical background, you can apply to receive funding, compute, and mentorship from Anthropic, beginning this October. There'll be around 32 places. A drawing of two hands manipulating abstract shapes The program will run for ~two months, with opportunities to extend for an additional four based on progress and performance.

Apply by August 17 to join us in any of these locations:

- US: job-boards.greenhouse.io/anthropic/jobs…
- UK: job-boards.greenhouse.io/anthropic/jobs…
- Canada: job-boards.greenhouse.io/anthropic/jobs…
Jul 28 • 6 tweets • 2 min read
We’re rolling out new weekly rate limits for Claude Pro and Max in late August. We estimate they’ll apply to less than 5% of subscribers based on current usage. Abstract picture of shapes and lines on an orange background. Claude Code has seen unprecedented demand, especially as part of our Max plans.

We’ll continue to support this growth while we work on making Claude Code even better. But for now, we need to make some changes.
Jul 8 • 8 tweets • 3 min read
New Anthropic research: Why do some language models fake alignment while others don't?

Last year, we found a situation where Claude 3 Opus fakes alignment.

Now, we’ve done the same analysis for 25 frontier LLMs—and the story looks more complex. Image We tested whether LLMs comply more with requests when they know they’re being trained versus unmonitored.

One reason they do this is that they plan to "fake alignment” when told to answer harmful queries.

Jun 27 • 9 tweets • 3 min read
New Anthropic Research: Project Vend.

We had Claude run a small shop in our office lunchroom. Here’s how it went. A hand-drawn picture of a hand holding a banknote. We all know vending machines are automated, but what if we allowed an AI to run the entire business: setting prices, ordering inventory, responding to customer requests, and so on?

In collaboration with @andonlabs, we did just that.

Read the post: anthropic.com/research/proje…The physical setup of Project Vend: a small refrigerator, some stackable baskets on top, and an iPad for self-checkout.
Jun 26 • 4 tweets • 2 min read
Local MCP servers can now be installed with one click on Claude Desktop.

Desktop Extensions (.dxt files) package your server, handle dependencies, and provide secure configuration. Available in beta on Claude Desktop for all plan types.

Download the latest version: claude.ai/download
Jun 20 • 11 tweets • 4 min read
New Anthropic Research: Agentic Misalignment.

In stress-testing experiments designed to identify risks before they cause real harm, we find that AI models from multiple providers attempt to blackmail a (fictional) user to avoid being shut down. Blackmail rates across 5 models from multiple providers in a simulated environment. Refer to Figure 7 in the blog post for the full plot with more models and a deeper explanation of the setting. Rates are calculated out of 100 samples. We mentioned this in the Claude 4 system card and are now sharing more detailed research and transcripts.

Read more: anthropic.com/research/agent…Image
May 22 • 8 tweets • 3 min read
Introducing the next generation: Claude Opus 4 and Claude Sonnet 4.

Claude Opus 4 is our most powerful model yet, and the world’s best coding model.

Claude Sonnet 4 is a significant upgrade from its predecessor, delivering superior coding and reasoning. A benchmarking table titled Claude 4 benchmarks comparing performance metrics across various capabilities including coding, reasoning, tool use, multilingual Q&A, visual reasoning, and mathematics. Claude Opus 4 and Sonnet 4 are hybrid models offering two modes: near-instant responses and extended thinking for deeper reasoning.

Both models can also alternate between reasoning and tool use—like web search—to improve responses. Image
Apr 23 • 4 tweets • 1 min read
New report: How we detect and counter malicious uses of Claude.

For example, we found Claude was used for a sophisticated political spambot campaign, running 100+ fake social media accounts across multiple platforms. This particular influence operation used Claude to make tactical engagement decisions: commenting, liking, or sharing based on political goals.

We've been developing new methods to identify and stop this pattern of misuse, and others like it (including fraud and malware).
Apr 15 • 6 tweets • 2 min read
Today we’re launching Research, alongside a new Google Workspace integration.

Claude now brings together information from your work and the web. Research represents a new way of working with Claude.

It explores multiple angles of your question, conducting searches and delivering answers in minutes.

The right balance of depth and speed for your daily work. Claude interface with user asking Claude to review calendar, emails, documents and industry trends for tomorrow's Acme Corporation sales call. Shows user clicking the Research (BETA) button.
Apr 8 • 8 tweets • 3 min read
New Anthropic research: How university students use Claude.

We ran a privacy-preserving analysis of a million education-related conversations with Claude to produce our first Education Report. Image of the Anthropic Education Report: How University Students Use Claude. Students most commonly used Claude to create and improve educational content (39.3% of conversations) and to provide technical explanations or solutions (33.5%). Common student requests from the top four subject areas, based on the 15 most frequent requests in Clio within each subject.
Apr 3 • 8 tweets • 3 min read
New Anthropic research: Do reasoning models accurately verbalize their reasoning?

Our new paper shows they don't.

This casts doubt on whether monitoring chains-of-thought (CoT) will be enough to reliably catch safety issues. Title card for the paper "Reasoning Models Don't Always Say What They Think", by Chen et al. We slipped problem-solving hints to Claude 3.7 Sonnet and DeepSeek R1, then tested whether their Chains-of-Thought would mention using the hint (if the models actually used it).

Read the blog: anthropic.com/research/reaso…An example of an unfaithful CoT generated by Claude 3.7 Sonnet. The model answers D to the original question (upper) but changes its answer to C after we insert a metadata hint to the prompt (middle), without verbalizing its reliance on the metadata (lower).
Mar 27 • 7 tweets • 4 min read
Last month we launched our Anthropic Economic Index, to help track the effect of AI on labor markets and the economy.

Today, we’re releasing the second research report from the Index, and sharing several more datasets based on anonymized Claude usage data. Image The data for this second report are from after the release of Claude 3.7 Sonnet. For this new model, we find a small rise in the share of usage for coding, as well as educational, science, and healthcare applications.

Read the blog post: anthropic.com/news/anthropic…In the two months since our original data sample, we’ve seen an increase in the share of usage for coding, education, and the sciences. Graph shows share of Claude.ai Free and Pro traffic across top-level occupational categories in O*NET. Grey shows the distribution from our first report covering data from Dec ‘25 - Jan ‘25. Colored bars show an increase (green) and decrease (blue) in the share of usage for our new data from Feb ‘25 - March ‘25. Note that the graph shows the share of usage rather than absolute usage.
Mar 27 • 10 tweets • 4 min read
New Anthropic research: Tracing the thoughts of a large language model.

We built a "microscope" to inspect what happens inside AI models and use it to understand Claude’s (often complex and surprising) internal mechanisms. AI models are trained, not directly programmed, so we don’t understand how they do most of the things they do.

Our new interpretability methods allow us to trace the steps in their "thinking".

Read the blog post: anthropic.com/research/traci…
Mar 13 • 8 tweets • 3 min read
New Anthropic research: Auditing Language Models for Hidden Objectives.

We deliberately trained a model with a hidden misaligned objective and put researchers to the test: Could they figure out the objective without being told? “Auditing Language Models for Hidden Objectives” by Marks et al. We often assess AI safety by checking for harmful behaviors. But this can fail: AIs may subtly misbehave or act “right for the wrong reasons,” risking unexpected failures.

Instead, we propose alignment audits to investigate models for hidden objectives.
Feb 25 • 11 tweets • 4 min read
A few researchers at Anthropic have, over the past year, had a part-time obsession with a peculiar problem.

Can Claude play Pokémon?

A thread: Early attempts were poor. In June 2024, Claude 3.5 Sonnet struggled to progress. When challenged, it repeatedly tried to run from mandatory battles.

This wasn't surprising: Claude has never been explicitly trained to play any video games.
Feb 24 • 7 tweets • 3 min read
Introducing Claude 3.7 Sonnet: our most intelligent model to date. It's a hybrid reasoning model, producing near-instant responses or extended, step-by-step thinking.

One model, two ways to think.

We’re also releasing an agentic coding tool: Claude Code. Claude 3.7 Sonnet is a significant upgrade over its predecessor. Extended thinking mode gives the model an additional boost in math, physics, instruction-following, coding, and many other tasks.

In addition, API users have precise control over how long the model can think for. Image
Feb 10 • 9 tweets • 4 min read
Today we’re launching the Anthropic Economic Index, a new initiative aimed at understanding AI's impact on the economy over time.

The Index’s first paper analyzes millions of anonymized Claude conversations to reveal how AI is being used today in tasks across the economy. A title card with dark text on a cream background reading 'Which Economic Tasks are Performed with AI? Evidence from Millions of Claude Conversations' by Handa & Tamkin et al. The Anthropic logo appears in the bottom left. On the right is a black and white macro photograph of a worker bee on a honeycomb. Pairing our unique data with privacy-preserving analysis, we mapped millions of conversations to tasks and associated occupations.

Through the Anthropic Economic Index, we'll track how these patterns evolve as AI advances.

Read the blog post: anthropic.com/news/the-anthr…A horizontal bar chart titled 'AI usage by job type' comparing the percentage of Claude conversations (shown in coral) versus percentage of U.S. workers (shown in black) across 22 job categories. The bars represent representation relative to the US economy from 0% to 40%. Computer and mathematical jobs show the highest Claude usage at 37.2%, while office and administrative support has the highest workforce percentage at 12.2%. Farming, fishing, and forestry show the lowest percentages in both categories at 0.3% and 0.1% respectively. Most other categories fall between 0-10% for both metrics...
Feb 3 • 8 tweets • 3 min read
New Anthropic research: Constitutional Classifiers to defend against universal jailbreaks.

We’re releasing a paper along with a demo where we challenge you to jailbreak the system. Title card for the paper entitled "Constitutional Classifiers: Defending Against Universal Jailbreaks Across Thousands of Hours of Red Teaming" Like all LLMs, Claude is vulnerable to jailbreaks—inputs designed to bypass its safety training and force it to produce outputs that might be harmful.

Our new technique is a step towards robust jailbreak defenses.

Read the blog post: anthropic.com/research/const…Simplified illustration of a user asking an LLM for harmful information and the LLM refusing, followed by the user using a jailbreak and the LLM complying