Aoki, also known as Aokigahara, is a data-stealing malware that has added credentials for AI tools to its growing list of targets. Recent versions seek local files and session data linked to ChatGPT, Claude, OpenAI Codex, Cursor, Codeium and Gemini. The change raises the risk for developers who use AI assistants alongside source-code repositories, cloud accounts and other work systems. Unlike a flaw in an AI service, Aoki’s activity focuses on data stored on an infected computer. A stolen key or token may let an intruder use an account as its owner, run up paid usage or reach services connected to that account. Earlier Aoki samples did not search for AI-related files, suggesting this is a newer addition to the malware’s collection rules. Researchers from Symantec’s Threat Hunter team identified the expanded targeting. Symantec’s Threat Hunter team analysis found that newer Aoki versions search for files linked to Claude, ChatGPT and Codex, as well as tokens used by services such as GitHub Copilot, Cursor, Google Gemini, JetBrains AI Assistant and Kiro. The Symantec Aoki threat report also describes the malware’s wider reach across browsers, wallets and messaging apps. The files named in the report include .claude.json , Claude desktop cookies, ChatGPT cookies, and Codex config.toml and .env files. These can contain login sessions, settings or secrets used by local tools. Aoki also targets Cursor’s local state database, which may hold chat context and code snippets. Symantec says the main focus is credentials and session data, rather than source code itself. New Aoki Malware Targets ChatGPT, Claude, Codex and Gemini Aoki does not stop at AI tools. It can collect browser cookies, saved passwords, autofill details, credit-card data and Google tokens from a broad range of browsers. Its other targets include cryptocurrency wallet files and extensions, password managers, VPN and file-transfer applications, gaming services, and Telegram and Discord data. It also records which security products are installed on the computer. That broad collection can turn one infected device into a source of access to several accounts. For developers, a stolen token may expose more than an AI subscription: connected repositories, cloud services and continuous integration and delivery pipelines could also be at risk. The growing focus on locally stored agent data has been seen in other reporting on AI coding agent data theft , though that coverage describes other malware families, not Aoki. Aoki sends collected information to a command-and-control server using a /gate endpoint. Symantec says newer samples used several domains, while older samples used an IP address and port 8080 over unencrypted HTTP. The report does not name a threat group behind Aoki, and it does not link the malware to a known family. Its name comes from “Aokigahara,” a string used during the malware’s decryption process. The theft of AI credentials can also lead to direct costs. An intruder may use a victim’s account quota or resell access; abusing paid AI services in this way is often called LLMjacking. A separate report on stolen AWS keys used for LLMjacking shows how compromised cloud credentials can be used to run paid AI models at the account owner’s expense. Aoki’s reported focus, however, is stealing data from infected devices. This pattern is not limited to Aoki. Cyber Security News has reported on a fake Gemini npm package that targeted credentials and local data from several AI development tools. The incidents underline why developers should treat AI configuration files and local session data as sensitive, much like cloud keys and source-control credentials. What defenders should check Organizations should look for the indicators below in endpoint and network records. A match is a reason to investigate, not proof on its own that a device is infected. If a system is suspected of compromise, teams should use a clean device to revoke active AI sessions, rotate exposed API keys and related credentials, and review account activity. They should also check connected repositories and cloud services for unexpected use. These steps matter because multi-factor authentication may not stop an attacker from reusing a token already stolen from a device. Developers should avoid saving long-lived secrets in files that are accessible to local applications, limit tokens to the permissions and time needed, and remove or replace exposed keys. Security teams can monitor for unexpected access to AI tool folders and unusual outbound connections, then compare findings with the indicators below. Symantec identifies the family as Infostealer.Aoki. Indicators of compromise (IoCs):- Type Indicator Reported label or note File hash 752c8e64af60960b85cf00e4bd34eb459e24b7939464f718ec18494df43b4b69 Infostealer.Aoki File hash bf037920d39b85572ec7e53b51f09c28198c2c0afbc7c7d6cf639b1f047caebf Infostealer.Aoki File hash 0adc8e83b2416c275b1283916d70568faddeb6d718eb45f9addfb0938db245b3 Infostealer.Aoki File hash 6655b89b9885483d459c73f2a52e3cec121de5c5956172a8676a0f4c2de94999 Infostealer.Aoki File hash 780bbdbb119befd4eb16e0c2ec0c64e7bedc1b53ce4f85804aea091a6e3af519 Infostealer.Aoki File hash 2fb80e2221bd304e7cd5ac04169e71807ad92d19836f1aaf7885295cb90b2d94 Infostealer.Aoki File hash 73dbac0907e8a092cd978f2787ab1994946f5da1e08f2bdbd05ea52e80cdb896 Infostealer.Aoki File hash 07bc2e3a1d14743cc44f1dc4c78d15d0f2b458b39fb48422b13df40fc662ba21 Infostealer.Aoki File hash c533570fd7cd7eaf5478317b03ea5750d03797aa4d11ebd96555b6cb708c63f2 Infostealer.Aoki File hash 6350dfee6975a9b159c5a57a3e97f757b0b02fd2dc708016fa6d7270e720c865 Infostealer.Aoki File hash 36af3acc78058023cdcf9307bc15be640be9425de97e574fdaf42d03d308725b Infostealer.Aoki File hash 039240551f88b0064b2615dffdc1321b1e7c58e3f617ac138a65a6d59400b331 Infostealer.Aoki File hash 0e38bd3c30a4ca08641240ce95b5b6ba593e131213ca1b2a28da4e0385fcabc4 Infostealer.Aoki File hash 3b4547f0d7549274f366cd9568b3d302c1e80e1b89359da681beb47260ec7975 Infostealer.Aoki File hash b590235493487c51f7050b341c9b2450e4f761ad6055207e4f6daf0cecae271c Infostealer.Aoki File hash e374f9ff5094f728528efa566c3de1b7fea7d95f497242a01fffa6c70b5eb768 Infostealer.Aoki File hash f779314b1a5a0b297fd4e2a5f9c1e40d387b3e7c401d42324b556a70f813840e Infostealer.Aoki Domain dobeqnesmot[.]cc Network IOC Domain hjazfioe[.]surf Network IOC Domain neihite[.]cc Network IOC URL hxxps://dobeqnesmot[.]cc/gate C&C endpoint URL hxxps://hjazfioe[.]surf/gate C&C endpoint URL hxxps://neihite[.]cc/gate C&C endpoint Note:   IP addresses and domains are intentionally defanged (e.g.,  [.] ) to prevent accidental resolution or hyperlinking. Re-fang only within controlled threat intelligence platforms such as MISP, VirusTotal, or your SIEM . Update defenses against new malware as soon as it appears. Integrate TI Feeds with data from 16K SOCs . The post New Aoki Malware Targets ChatGPT, Claude, Codex and Gemini to Steal Credentials appeared first on Cyber Security News .

Key findings The Aoki (aka Aokigahara) infostealer is now stealing credential and session files used by AI coding assistants, including Claude, OpenAI's Codex, Cursor, Codeium and Gemini. Earlier versions of the malware did not target AI-related files, which indicates the capability is a recent addition. Stolen application programming interface (API) keys and tokens can allow attackers to carry out activity under a legitimate user's identity and API quota.

はじめに 2026年10月現在、AIコーディングツールは「使うかどうか」ではなく「どれを、どういう設定で、どう組み合わせるか」を考える段階に入っています。本記事では主要5ツール — Claude Code / GitHub Copilot / Cursor / Winds...

In this article, I will cover the Best Windsurf Alternatives for AI programming Assistant 2026. I will be looking at solutions that provide independence, privacy and flexibility. I will cover Cursor and GitHub Copilot and Codeium and Claude Code. All of the above solutions provide advances in Agent-based programming and integrate with IDEs. They provide examples of how programming Assistants have advanced to the enterprise level. All of the above solutions should be evaluated by programmers working independently and in smaller organizations as well as larger and more established organizations.

AI 코딩 도구를 쓰다 보면 어느 순간 한계가 느껴집니다. Windsurf는 에이전트 방식으로 개발 흐름을 이해하는 게 특징이지만, 가격이나 안정성 면에서 아쉬움을 느끼는 분들이 꽤 있습니다. 그래서 실제 개발자들이 많이 옮겨 간 대안 5곳을 직접 비교해 정리했습니다.

> **TL;DR:** Claude Code is a CLI tool that reads your entire codebase and makes coordinated multi-file changes. It cut my feature development time from 4-6 hours to 1-2 hours. Not perfect for large codebases or visual design, but transformative for real development work. ## My Background With AI Coding Tools I have tried everything. GitHub Copilot for a year. Cursor for six months. ChatGPT for code generation since the GPT-4 launch. Amazon CodeWhisperer. Codeium. Tabnine. I have a strong opinion on what works and what is marketing hype. Claude Code is different from all of them. Not because it writes better code — although it does — but because it fundamentally changes what "using an AI coding tool" means. This is not autocomplete on steroids. This is a developer sitting next to you who can read your entire codebase, understand your architecture, and make meaningful changes across multiple files. Let me explain. ## What Claude Code Actually Is Claude Code is a CLI tool from Anthropic. You run it in your terminal, inside your project directory. It can read your files, understand your project structure, run commands, and make edits. The key difference from other tools: it operates on your real codebase, not a chat window where you copy-paste code back and forth. When I say "fix the authentication bug in the user service," Claude Code will: 1. Read the relevant files 2. Understand the codebase context 3. Find the bug 4. Edit the files to fix it 5. Run the tests to verify All in one go. No copy-pasting. No explaining your file structure. No "here is the fix, now go apply it manually." ## What Works Incredibly Well ### Multi-File Edits This is Claude Code's killer feature. Most AI coding tools work within a single file. Claude Code works across your entire project. When I ask it to "add a new API endpoint with validation, database model, and tests," it creates the route file, the model, the validation schema, the tests, and updates the barrel exports — all coordinated and consistent. I recently used it to add a complete lead capture system to a website. Contact form, API route, JSON storage, Telegram notifications, form validation — across 8 files, all in one session. Would have taken me half a day. Claude Code did it in 15 minutes. ### Understanding Existing Code Claude Code reads your codebase before making changes. It does not guess at your patterns or conventions — it looks at your existing code and follows them. If your project uses a specific error handling pattern, Claude Code will use the same pattern. If you have a constants file, it will add new constants there instead of hardcoding values. This sounds simple, but it is a massive quality improvement over tools that generate code in isolation. The output feels like it belongs in your project because Claude Code actually read your project before writing. ### Explaining and Debugging Sometimes I do not want code written. I want to understand why something is broken. Claude Code is exceptional at reading a stack trace, finding the root cause across multiple files, and explaining the issue clearly. Last week, a production deploy failed with a cryptic webpack error. I pasted the error into Claude Code and asked it to debug. It traced the issue to a circular dependency between two modules, explained exactly why it caused the build failure, and fixed it. What would have been an hour of git bisect and console logging was solved in two minutes. ### Terminal Integration Because Claude Code runs in the terminal, it can run your actual commands. npm test, git status, database migrations — it can execute and read the output. When I say "run the tests and fix any failures," it runs the test suite, reads the output, identifies the failing tests, fixes the code, and re-runs the tests. This feedback loop is incredibly productive. Write code → run tests → fix failures → verify — all without leaving the AI conversation. ## What Does Not Work (Yet) ### Large Codebase Navigation On very large projects (50K+ lines), Claude Code sometimes struggles to find the right files. It can read any file you point it to, but discovering which files are relevant in a massive codebase requires some guidance. I have learned to be specific: "look at the auth middleware in src/middleware/auth.ts" rather than "fix the auth bug." ### Complex Refactoring For straightforward refactoring — rename a variable, extract a function, move a file — Claude Code is excellent. For complex architectural refactoring — changing a monolith to microservices, rewriting a state management approach — it needs more hand-holding. The AI works best when the changes are well-scoped. ### Frontend Visual Design Claude Code writes functional frontend code, but it does not "see" the result. If I say "make the button look better," it will change CSS properties, but it cannot evaluate whether the result actually looks better. For visual work, I still use the browser and make adjustments manually. ### Rate Limits During heavy usage sessions, you hit rate limits. This is the reality of API-based tools — there is a cost per token, and extended sessions with large codebases burn through context quickly. I have learned to break my work into focused sessions rather than trying to do everything in one conversation. ## How It Changed My Workflow ### Before Claude Code My typical day looked like: 1. Read the requirement 2. Plan the implementation in my head 3. Write code file by file 4. Copy-paste between files to maintain consistency 5. Write tests 6. Debug failures 7. Iterate Total for a medium feature: 4-6 hours. ### After Claude Code 1. Read the requirement 2. Tell Claude Code what I want, with context on constraints and patterns 3. Review the changes it makes 4. Run tests, have Claude Code fix any issues 5. Make manual adjustments for edge cases Total for a medium feature: 1-2 hours. The time savings are real, but the bigger change is cognitive. I spend less mental energy on boilerplate and mechanical coding, and more on architecture, design decisions, and code review. I am a better developer because I can focus on the hard problems. ## Claude Code vs The Competition ### vs GitHub Copilot Copilot is autocomplete. Claude Code is a collaborator. Copilot suggests the next line. Claude Code understands the task and builds the solution. They are not competing products — they solve different problems. I use both: Copilot for inline suggestions while typing, Claude Code for feature-level work. ### vs Cursor Cursor is closer to Claude Code in ambition. It understands project context and can make multi-file edits. The main differences: Claude Code runs in the terminal (I prefer this — no IDE lock-in), and Claude's model is significantly better at understanding complex codebases and Indian English instructions. Cursor has better visual integration since it is a full IDE. Claude Code has better reasoning and more reliable code generation. If you live in VS Code, Cursor is convenient. If you live in the terminal, Claude Code is superior. ### vs ChatGPT ChatGPT for coding is a chat window. You paste code in, get code out. There is no project awareness, no file editing, no command execution. For quick questions and isolated code snippets, ChatGPT is fine. For real development work, it is not in the same category as Claude Code. ## Tips for Getting the Most Out of Claude Code 1. **Be specific about context.** Tell it which files matter, which patterns to follow, which constraints exist. The more context, the better the output. 2. **Work in focused sessions.** Instead of one massive conversation, break work into feature-sized sessions. This keeps context sharp and avoids rate limits. 3. **Review everything.** Claude Code is good, but it is not infallible. Read the diffs. Understand the changes. Do not blindly accept. 4. **Use it for exploration.** When you are unsure how to approach a problem, ask Claude Code to analyze the codebase and suggest approaches. Even if you do not use its code, the analysis is valuable. 5. **Set up your CLAUDE.md.** The CLAUDE.md file in your project root tells Claude Code about your conventions, patterns, and preferences. A well-written CLAUDE.md dramatically improves output quality. Claude Code uses MCP Protocol under the hood for tool connections. I've used it to build systems that saved clients ₹85K/month on AI costs and to create n8n automation workflows in a fraction of the usual time. ## The Bottom Line Claude Code is not a replacement for developers. It is a force multiplier. It handles the mechanical, repetitive, cross-file work that eats up development time, freeing me to focus on the problems that actually require human judgment. Is it perfect? No. Large codebase navigation needs improvement, rate limits are frustrating during intensive sessions, and it cannot evaluate visual design. But the core capability — understanding a codebase and making coordinated, multi-file changes based on natural language instructions — is genuinely transformative. I build faster, ship more, and spend more time on the interesting problems. For a developer who values productivity and is comfortable reviewing AI-generated code, Claude Code is the most impactful tool I have adopted in years. If you want to see how Claude Code can accelerate your development workflow, book a session and I will walk you through a live demo. _I write about automation and the systems I actually run at architmittal.com. Originally published there._

Hey fellow developers! 👋 VS Code recently introduced a fun experimental feature: an interactive "pet" companion that sits above the chat input and reacts to AI agent activity. The only catch? **It currently only reacts to GitHub Copilot.** I believe this awesome feature should also recognize free and open-source AI coding assistants like **Cline, Kilo Code, Tabby, Codeium, Roo Code** , and others. I've opened an official feature request in the VS Code repository, and it has been accepted as a "Backlog Candidate". However, there's a rule: **it needs 20 upvotes (👍) within 60 days** to stay open and be considered for development. If you use any alternative AI coding assistant and want to see native, playful support for it in VS Code, please take 5 seconds to drop a 👍 reaction on the main post of the issue! 🔗 **Link to upvote:** https://github.com/microsoft/vscode/issues/335649 _(Note: Only the emoji reaction on the very first post counts towards the VS Code bot's total. Comments like "+1" do not work)._ Let's make VS Code more inclusive for all open-source tools! Thanks for your support. 🚀

Discover codeium.vim on Open Source Projects

OpenAI is in talks to drop $3 billion to purchase Windsurf, which sounds like a rival of Pacific Sunwear at your local mall, but is actually an AI coding assistant formerly known as Codeium. The potential deal is a signal that CEO Sam Altman is preparing to stake his claim in the burgeoning field…

OpenAI has agreed to buy Windsurf, an artificial intelligence-assisted coding tool formerly known as Codeium, for about $3 billion, according to people familiar with the matter, marking the ChatGPT ma...