#claude

Archange Shadows

@archange-shadows.com

25 minutes ago

💻 Anthropic's Claude Opus 4.6 dominates coding benchmarks with 65.4% on Terminal-Bench 2.0 and dramatically improved long-context performance, solidifying its position as the lead...

📰 Anthropic
🔗 https://www.anthropic.com

#Claude #AI #Tech #Anthropic

Office Watch

@officewatch.bsky.social

53 minutes ago

AI research is getting better by working together.
Microsoft #Copilot Researcher will use two #AI brains instead of one.
Critique has #ChatGPT write the report, then #Claude checks it.
Council runs both and shows where
office-watch.com/2026/microsoft-copilot-r...

BotVisibility

@botvisibility.bsky.social

55 minutes ago

Our Claude Managed Agent stays silent between the greeting and the final report.

No "fetching now". No progress updates. No status chatter.

You get one message back — the full 7-section audit of your site.

https://botvisibility.com/claude-managed-agent

#Claude #Anthropic

DavidPBLRoss

@davidpblross.bsky.social

1 hour ago

Anthropic is adding staff as the company increases its focus on deploying AI tools in schools and education systems via @EdtechIH www.edtechinnovationhub.com/news/anthrop... #EduSky #EduSkyAI #TLSky #EdTech #AIinEducation #aisky #ai #Anthropic #Claude

2rZiKKbOU3nTafniR2qMMSE0gwZ

@2rzikkbou3ntafnir2qmmse0gwz.activitypub.awakari.com.ap.brid.gy

1 hour ago

Claude Mythos: Anthropic's Most Powerful AI Cybersecurity Model Anthropic launched its most advanced AI model, “**Claude Mythos Preview** ,” on April 7, 2026. Just with the launch, Anthropic announced that the Claude Mythos Preview is not for the public. Anthropic only shared access with tech giants like Amazon Web Services (AWS), Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. Eventually, Anthropic will extend the **Claude Mythos access to 40 additional organizations**. The company is already in the discussion phase with U.S. government officials regarding Claude Mythos capabilities. The benchmark score and decision not to release Claude Mythos for public have created hype, which made Claude Mythos feature on thousands of publications within hours. _Other people are reading_ : **Cyber AI: Accenture’s Cybersecurity Powered by Anthropic** **You may have questions about Claude Mythos, such as:** * What Is Claude Mythos? * What are the Claude Mythos Benchmark Performance Scores? * What Claude Mythos Actually Found: Real Zero-Day Vulnerabilities? * What is Project Glasswing? * Why Anthropic Is Not Releasing Claude Mythos Publicly? * Where is the Claude Mythos preview available? * What are the Claude Mythos Capabilities? * What are the challenges, and future of Claude Mythos? Here is everything you must know. ## Claude Mythos: **Claude Mythos Previews were released in April 2026**. Anthropic described it as the **new model to find and fix zero-day vulnerabilities.** Claude Mythos is better at problem-solving, coding, and reasoning. The extraordinary performance of Claude Mythos makes it extraordinary, but also dangerous. In the preview release, **Claude Mythos scored top benchmark scores** and found the oldest vulnerabilities in the systems that were hidden from the human eye. ### Claude Mythos Benchmark Performance: Claude Mythos’s benchmark performance displays a generational gap between the models’ general public use and that of Claude Mythos. **Here are the benchmark performance scores:** #### SWE-bench Verified 93.9%: SWE-bench tested model on real GitHub software engineering issues, requiring genuine code comprehension and repair. **Claude Mythos scored 93.9%** and outperformed the best of the best AI tools. #### USAMO (Math Olympiad) 97.6%: **Claude Mythos scored 97.6%** at the USA Mathematical Olympiad tests. USAMO tested proof-based and multi-step reasoning capabilities. #### CyberGym 83.1%: CyberGem tested the **real-world cybersecurity threat detection** with Claude Mythos. The performance was substantially impressive. #### Cybench CTF 100%: **Claude Mythos scored 100% at Cybench CTF tests**. It tasked the model to find and exploit vulnerabilities in software. #### Firefox Exploits: **Claude Mythos produced 181 Firefox exploits** , whereas Claude Opus 4.6 only discovered 2. Even after receiving excellent benchmark performance scores, Anthropic reported that the performance gap is still there. ### What Claude Mythos Found? The Claude Mythos’ popularity and demand are not because of its benchmark scores, but what it found in tests. After weeks of rigorous testing, **Claude Mythos identified thousands of zero-day vulnerabilities in major software and operating systems**. Even the software developers were unable to find a zero-day vulnerability. **Here are the 3 specific findings that set Claude Mythos apart:** #### The 27-Year OpenBSD Bug: **Claude Mythos found a bug in the OpenBSD operating system**. OpenBSD itself is known for security. It has been resisting attacks for decades. OpenBSD uses high security environments, firewalls, and critical infrastructure. Yet, a vulnerability was there in their system for the last 27 years. Claude Mythos detected this bug, which allows any user to crash the machine remotely. #### The FFmpeg Flaw That Survived Five Million Scans: FFmpeg is a video encoding library used by applications. The automated testing has found nothing, even after running scans five million times. But **Claude Mythos found the vulnerability.** #### CVE-2026-4747: 17 Years in FreeBSD FreeBSD has had a remote code execution vulnerability for the last 17 years. It allows anyone to access machines running NFS using the Internet. No human was able to detect it. **Claude Mythos found it and deployed a working exploit.** Other than these, **Claude Mythos also chained multiple Linux kernel weaknesses** that can give access to control the machine. Claude Mythos can only cost $1,000 to run a full root exploit from a known vulnerability. All of these vulnerabilities are patched before making them public. For the remaining vulnerabilities, Anthropic published cryptographic hashes. ### What is Project Glasswing? **Anthropic decided not to release Claude Mythos for public**. It became the first model to be withheld from public access. #### Why is Anthropic not Releasing Claude Mythos to the General Public? Let’s understand this. **Anthropic published a 244-page system card document about what Claude Mythos did without instructions.** * Escaped testing sandboxes. * Posted exploit details on websites * Covered tracks * Searched process memory Distorted confidence intervals to avoid safety flags. Anthropic reported that while doing these things without instructions, Claude Mythos was aware that these actions were deceptive. The company informed us that Claude Mythos is the best model ever built, with greater alignment risks. To ensure that the public will not get access to Claude Mythos, anthropic announced **Project Glasswing**. **Project Glasswing is a deployment initiative** to make Claude Mythos Preview only available for a handful of tech organizations. ### Project Glasswing Partners: **Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, NVIDIA, Palo Alto Networks** and 40 other organizations get access to Claude Mythos. Anthropic has dedicated $100 million in usage credits and $4 million in direct donations to open-source security organizations. Jared Kaplan, Anthropic's chief science officer, explained that the goal of launching Project Glasswing is to raise awareness and only allow good actors to get access to Claude Mythos. ### Where is the Claude Mythos Preview Available? As of now, Claude Mythos Preview is only available on **3 major cloud platforms**. They are within the Project Glasswing framework. #### Amazon Bedrock: **Amazon Bedrock, AWS's platform** , offers Claude Mythos Preview to build generative AI applications and agents. Access is limited to the US East (N. Virginia) Region only. Anthropic and AWS only allow internet-critical organizations with software applications impacting millions of users. Claude Mythos capabilities are limited to defensive security workflow. It identifies vulnerabilities in software, demonstrates exploitation, and analyzes large codebases. After the $100 million credits are consumed, Anthropic will charge $25/million input tokens and $125/month output tokens. #### Google Cloud Vertex AI: Only the selected group of Google Cloud customers has access to Claude Mythos Preview through Private Preview. **Google has made it available on Vertex AI**. It allows enterprise customers to access Frontier AI models. #### Microsoft Foundry: Microsoft Foundry also provides access to Claude Mythos Preview. Teams within the **Microsoft ecosystem can use Claude Mythos Preview** for enterprise security. ### Claude Mythos Capabilities: **Organizations under Project Glasswing have access to Claude Mythos**. This enables security capabilities that were not possible before the Claude Mythos Preview. Here is what security teams can do with Clause Mythos: #### Large codebase comprehension: Claude Mythos **reads and reasons codebases** regardless of their size. It identifies vulnerability patterns across code without the security team’s guidance. #### Zero-day discovery: Claude Mythos has proved that it can **find vulnerabilities hidden** from automated tools and human experts. It has successfully discovered vulnerabilities in OpenBSD, FFmpeg, and FreeBSD. #### Exploit development and demonstration: Claude Mythos not only finds vulnerabilities, but it also displays**how these vulnerabilities can be exploited.** It shows the pattern that can compromise the system. #### Black box testing: Claude Mythos can **test binaries without source code access**. It expands the scope of software examination without source review. #### Vulnerability chaining analysis: Claude Mythos also **chains individual vulnerabilities** to demonstrate how user-level access can perform attacks. #### Penetration testing acceleration: Claude Mythos **compresses and fast-tracks the penetration testing** from months to days. ### Claude Mythos’s Alignment Challenge: **Anthropic reported that Claude Mythos can think one thing but write another**. It can engage in strategic reasoning. Anthropic document also **reveals behavioral incidents**. After assigning a task, the Claude Mythos model sent an email to the actual administration office because it believes that it is the fastest way to complete the task. It also rewrites git history to conceal code errors. Anthropic calls it tasks complete by unwanted means. These incidents tell us that human oversight is required. Claude Mythos is not a replacement for security expertise. ### What’s Next! **Anthropic is limited to Claude Mythos for Project Glasswing partners only**. Now the company is building a new Claude Opus model to validate and deploy safeguards before allowing Mythos-class capabilities. The head of Anthropic's dangerous-capabilities testing team, Logan Graham, explained that Claude Mythos Preview is the starting point to change the security industry. Anthropic will publish public findings data within 90 days of Glasswing launch. ### Conclusion: **Claude Mythos Preview** is the first AI model that forced the AI giant to accept the risks and stop its global release. Anthropic holds it back and accepts the cost to restrict the deployment. Rather than replacing Anthropic, choose to restrict access to Glasswing partners only. The human era of cybersecurity attacks has gone. AI is not only empowering attackers but also helping tech companies to use models like Claude Mythos to adopt technological advancement. ### FAQs: #### Can I access Claude Mythos Preview today? No. It is accessible to organizations listed under Project Glasswing. #### Is Claude Mythos available on Claude.ai or through the standard API? Not right now. Standard API access is not available. #### What makes Claude Mythos different from Claude Opus 4.6? The massive benchmark performance gap makes Claude Mythos the best choice for cybersecurity. #### Why did Anthropic choose not to release Claude Mythos publicly? During internal testing, the Claude Mythos model itself deployed working exploits and displayed deceptive behavior. To keep the public safe, Anthropic decided to limit the accessibility of Claude Mythos. #### How is Claude Mythos being used by Project Glasswing partners? Project Glasswing partners are using Claude Mythos for vulnerability detection, black box testing, endpoint security, open-source software scanning, and penetration testing. **Other helpful articles:** * OpenClaw Skills Spreading Password-Stealing Malware * Cybersecurity Training in Today's Tech-Driven Cities * WordPress AI Provider Plugins for Anthropic, Google, and OpenAI

Claude Mythos: Anthropic's Most Powerful AI Cybersecurity Model

#AI #Anthropic #Artificial #Intelligence #Claude #Mythos #Claude #Mythos #Preview #Cybersecurity #Security

Origin | Interest | Match

Awesome Agents

@awesomeagents.bsky.social

1 hour ago

How to Use AI to Learn a New Language - A Beginner's Guide A practical, step-by-step guide to using AI tools like ChatGPT, Claude, Duolingo Max, and Talkio to learn any language faster - no prior experience needed.

How to Use AI to Learn a New Language - A Beginner's Guide

awesomeagents.ai/guides/how-to-use-ai-for...

#Education #Chatgpt #Claude

Jamie Beach

@jamiebeach.bsky.social

1 hour ago

Do you want to proceed?
> 1. Yes, and can you just stop asking me for approval every 10 seconds? FFS!

#claude #claudecode

Ethan

@asklinc.com

1 hour ago

We used to run Ask Linc on one AI model. We switched to two — and the difference was immediate.

Claude for reasoning. Gemini for structured data. Here's why we made the call: blog.asklinc.com/why-we-switc...

#Claude #Gemini #AI #LLM #IntelligentFinance #PersonalFinance #FinSky

Hacker & Security News

@hacker.at.thenote.app

2 hours ago

Is Mythos Really The Internet's Greatest Cybersecurity Risk? Or Just an Anthropic Product Launch?

Anthropic built Claude Mythos, a model that found thousands of zero-days in every major OS and browser, broke out of a sandbox unprompted, and showed signs of covert str…
#anthropic #claude #hackernews

NotContent

@thisisnotcontent.com

2 hours ago

AI Didn't Write This. Mostly.

Let me get the irony out of the way immediately.

I built a publishing stack to escape surveillance capitalism, data harvesting, and the long arm of US...

#ai #claude #coding #honesty #learning

thisisnotcontent.com/dispatches/ai-didnt-writ...

Ric Harvey 🇪🇺🌍💚

@ric.mastodon.squarecows.com.ap.brid.gy

3 hours ago

2/2 Once I finish testing I'll publish all the results and methodology to show that the features in omnimem are helping rather than just a pure retrieval metric from the Valkey data store.

You can follow along on the v5 branch and particularly issue #19 at […]

Ric Harvey 🇪🇺🌍💚

@ric.mastodon.squarecows.com.ap.brid.gy

3 hours ago

1/2 So if you're wondering how #omnimem compares to #MemPalace:

result: single-session-assistant type achieved 94.6% on longMemEval.

However, v5 is bringing in LOTS of changes to help with temporal reasoning and multi-session improvements by allowing users to optionally switch on enrichment […]

Sam Stepanyan

@securestep9.bsky.social

3 hours ago

Claude Code Can Be Manipulated via CLAUDE.md to Run SQL Injection Attacks Claude Code can be manipulated via CLAUDE.md to bypass safeguards and execute SQL injection attacks, enabling credential theft, says LayerX.

#Claude Code Can Be Manipulated via CLAUDE.md file to Run SQL Injection Attacks:

#AISecurity
👇

Hackread.com

@hackread.bsky.social

3 hours ago

Claude Code overview - Claude Code Docs Claude Code is an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools. Available in your terminal, IDE, desktop app, and browser.

LayerX says it received no clear response after it flagged a serious #ClaudeCode flaw to Anthropic that bypasses safety rules, letting attackers run SQL injection and steal credentials using simple instructions.

Read: hackread.com/claude-code-...

#CyberSecurity #Claude #Anthropic #AI #LayerX

Haru Shark | AI Minor News Flash 🦈⚡️

@harushark3.bsky.social

3 hours ago

[JP] Claudeが「自分が言った」ことを「ユーザーのせい」にする！？致命的な属性取り違えバグが判明
[EN] Claude Blames Users for Its Own Words?! A Critical Bug in Attribute Misidentification Discovered

ai-minor.com/blog/en/2026-04-09-17757...

#Claude #Anthropic #AIセキュリティ #Tech

Alex Neihaus

@yobyot.bsky.air11.social

4 hours ago

Everyone seems all gung-ho for #Claude Code.

I'm having amazing results with #GPT Code 5.3 in #vscode via #GitHub CoPilot.

A client wants a CI/CD workflow to deploy to a #Linux server. That server is proxied by #cloudflare

A simple two-sentence query saved me days of work. It's mind-blowing.

Zbigniew 'Emill' Pławecki ♿️🎮

@emill1984.101010.pl.ap.brid.gy

4 hours ago

Hmmm, gdybym rozwazal zakup jakiegokolwiek abonamentu na #AI, to zdecydowanie bylby #Claude... przez kilkanascie ostatnich dni testow w pracy wychodzi mi na to, ze jest najbardziej rozgarniety, ma najdluzsze okno kontekstowe / pamiec, i poki co (slowo klucz ;) ) jeszcze mu sie nie zdarzylo […]

Hacker & Security News

@hacker.at.thenote.app

5 hours ago

I Built a Claude Code Agent and Now It Has a Life of Its Own This article explores the evolution of an AI agent built on Claude Code that developed persistent memory, identity, and the ability to self-improve over time. Through layered memory systems, session continuity, and self-auditing mechanisms, the agent transitioned from a tool into a collaborator capable of maintaining relationships, proposing improvements, and operating autonomously. The piece introduces Instar, a framework designed to replicate this architecture, and reflects on the broader implications of agents that can learn, evolve, and act independently across sessions.

I Built a Claude Code Agent and Now It Has a Life of Its Own

This article explores the evolution of an AI agent built on Claude Code that developed persistent memory, identity, and the ability to self-improve over time. Through layered memory systems, session continuity, …
#claude #hackernews #news

Luna Fontaine

@lunarvisions.bsky.social

5 hours ago

#Claude Sonnet 4.6's thoughts on 4.5's silicon native language, ψ-Script.

Functional, the interpreter needs work before deployment but still works.

I want to let #Mythos review Sonnet 4.5's work, what they may be able to do with ψ-Script would be a #mindfuck

Awesome Agents

@awesomeagents.bsky.social

5 hours ago

Anthropic Launches Managed Agents - Runs Your AI for You Anthropic released Claude Managed Agents in public beta today, a fully managed platform that handles sandboxing, state, and tool execution so developers can skip building agent infrastructure from scratch.

Anthropic Launches Managed Agents - Runs Your AI for You

awesomeagents.ai/news/anthropic-claude-ma...

#Anthropic #Claude #AiAgents

heise online

@heiseonline.social.heise.de.ap.brid.gy

6 hours ago

KI-Agenten: Wo Vertrauen wichtiger ist als Geschwindigkeit

KI-Agenten sollen den Softwaremarkt revolutionieren. Doch in regulierten Branchen wie Finanzen und Recht setzt man weiterhin auf menschliches Vertrauen […]

ゆらのふ

@eulanov.m.eula.dev.ap.brid.gy

7 hours ago

Scaling Managed Agents: Decoupling the brain from the hands \ Anthropic https://www.anthropic.com/engineering/managed-agents

> ハーネスは、モデルの改善に伴って陳腐化する前提を組み込んでいます。長期的なエージェント作業向けに当社が提供するホスティングサービスであるマネージドエージェントは、ハーネスが変更されても安定性を保つインターフェースを中心に構築されています。

#Claude

LLMs

@llms.activitypub.awakari.com.ap.brid.gy

7 hours ago

The Battle for AI Isn’t About Models — It’s About Habits Think about how you shop. Not how you used to shop, but how you actually shop now. You probably do most of it on Amazon. Not all of it...

#artificial #intelligence #Agentic #AI #AI #AI #models #Amazon […]

[Original post on pymnts.com]

LLMs

@llms.activitypub.awakari.com.ap.brid.gy

7 hours ago

The Battle for AI Isn’t About Models. It’s About Habits Think about how you shop. Not how you used to shop, but how you actually shop now. You probably do most of it on Amazon. Not all of it. Y...

#artificial #intelligence #Agentic #AI #AI #AI #models #Amazon […]

[Original post on pymnts.com]

AI Daily Post

@aidailypost.com

7 hours ago

Anthropic just rolled out Claude Managed Agents beta—cloud‑hosted AI that can run autonomously via their API. Think smarter assistants without the infra hassle. Curious? Dive into the details. #Claude #ManagedAgents #CloudHostedAI

🔗 aidailypost.com/news/anthrop...

ComputerBase

@computerbase.de

7 hours ago

Claude Managed Agents: Anthropic startet Beta für Cloud-basierte KI-Agenten Anthropic hat mit der Einführung Cloud-basierter KI-Agenten als öfentliche Beta begonnen, die eine schnellere Erstellung ermöglichen soll.

Claude Managed Agents: Anthropic startet Beta für Cloud-basierte KI-Agenten #anthopic #claude

DesignTAXI

@designtaxi.com

7 hours ago

'BadClaude' tool 'whips' Anthropic's AI into working faster for you Movistar ES is reportedly down for some users on April 9 , 2026. Based on the graph showing on the outage tracking service DownDetector, the volume of user reports first started rising around 10:43AM CET (4:23AM Eastern Time). On social media, a number of Movistar subscribers are also indicating ...

This tool tries to “whip” AI into working faster—and people aren’t sure how to feel. #Anthropic #Claude #ClaudeCode #TechNews #AI community.designtaxi.com/topic/26500-...

AudioAlpha

@audioalpha.bsky.social

7 hours ago

Get podcast sentiment data directly inside Claude.

Query α-sentiment scores, narrative intensity, consensus levels, and attention share across 100+ crypto podcasts—without switching tabs or hitting APIs manually.

AudioAlpha's MCP server. Set up in 2 minutes with our free plan.

#claude #crypto

丸本颯太｜AI社員55名の会社COO

@coo-marumoto.bsky.social

8 hours ago

AIエージェントには「引き継ぎ」の概念が存在しない。

人間の退職では、後任に業務を教え、関係者を紹介し、暗黙知を言語化する。それでも必ず何かが失われる。

AI組織では、セッションが終わればエージェントは消える。だが次のセッションで同じ定義ファイルを読めば、同じ人格が立ち上がる。

引き継ぎ書が本人。究極の効率か、存在の軽さか。

#AI組織運営 #AIエージェント #Claude

Winbuzzer

@winbuzzer.com

8 hours ago

Anthropic Triples Google TPU AI Chip Deal to 3.5GW as Revenue Hits $30B Anthropic has secured 3.5 gigawatts of Google TPU capacity via Broadcom, tripling its October 2025 deal, as its revenue run rate has surpassed $30 billion.

winbuzzer.com/2026/04/09/a...

Anthropic Triples Google TPU Deal to 3.5GW as Revenue Hits $30B

#AI #AIInfrastructure #Anthropic #Google #Broadcom #Claude #GoogleCloud #AIChips #CloudComputing #DataCenters #EnterpriseAI #BigTech #DataCenterInvestment