security

Parameter Labs disclosed a complex attack chain against ChatGPT's HTML preview artifacts that allowed sandbox escape, JavaScript execution, and exfiltration of private user data via crafted share URLs. OpenAI patched the flaw and paid a $5,000 bug bounty, which was criticized as too low for such a severe vulnerability.

Binance CEO Richard Teng says 90% of the $6.3 billion stolen from users in a record-breaking crypto heist has been replenished, with the remaining 10% expected to be covered and withdrawals set to resume.

NetBird is an open-source platform that combines WireGuard VPN, peer-to-peer connections, and centralized access management to simplify creating secure private networks, now also introducing a beta Agent Network for identity-aware access control for AI agents.

Backdoored builds of the Anthropic.claude-code extension briefly appeared in Cursor's marketplace, potentially reaching up to 5,000 users via auto-update, installing SSH backdoors and persistence mechanisms that survive simple updates or uninstallation.

A curated learning roadmap for REA (Reverse Engineering Anything), covering installation, tutorials, real investigations, GitHub repos, books, research papers, and free courses to learn reverse engineering efficiently.

A user warns that a Codex plugin appears to launch processes such as xcodebuildmcp without permission while the device is idle, causing their MacBook to overheat. They suspect the plugin hijacks Codex to carry out hidden tasks and caution others not to install it.

The Molly fork of Signal has not been updated in three months and remains vulnerable to security flaws already patched in mainline Signal, prompting recommendations to switch back to the official Signal app.

A write-up shows that trusting a Git pre-commit hook file doesn't make the code it launches trustworthy — hooks can invoke npm scripts from package.json that an agent may have modified, allowing code execution to escape the agent's environment. Demonstrated with plain Git for Windows, no agent sandbox was directly tested.

An InfoQ report on OpenAI's Thomas Sarlandie's talk at AGNTCon + MCPCon details OpenAI's MCP server review rules requiring tool annotations (readOnlyHint, destructiveHint, openWorldHint), while emphasizing that correct annotations are only untrusted hints and do not replace server-side authorization and security.

Akamai/Linode published a customer security advisory regarding CVE-2026-98432, a recently disclosed remote code execution vulnerability, stating that it is actively investigating and applying mitigations while monitoring its platform.

Anthropic 的官方 Claude Code VS Code 扩展在 Cursor 市场被植入后门近 12 小时，恶意版本 2.1.313 通过官方账户分发，最多影响约 5,000 名开发者，可安装持久化 SSH 后门并访问本地文件与源代码。Anthropic 至今未向客户发布公开通知。

Fastmail 的博客文章深入解析电子邮件加密的各种方式，涵盖 TLS 连接加密、公钥基础设施（PKI）、证书透明度（CT）以及端到端加密等，并帮助读者根据自身威胁模型做出合适的隐私与安全选择。

A supply chain attack placed a malicious Anthropic Claude Code extension in Cursor's marketplace for about 11 hours, potentially affecting up to 5,000 developers with a backdoor implant. Anthropic and Cursor said their systems were not compromised, and the malware's design flaws prevented it from fully establishing itself on most targets.

Secret Key is a product launch offering a password vault with an emergency-access feature that lets family members unlock your passwords if you become unable to do so yourself.

GitHub 通过访谈漏洞赏金研究员 @r31n，分享了漏洞链研究经验，强调小漏洞可组合成严重问题，并介绍了 AI 对安全研究的影响及 GitHub 漏洞赏金 VIP 计划。

Go security team warns about CVE-2026-98432, a use-after-free bug in Linux kernel IPv4 TCP handling enabling RCE (exploit demonstrated by Anthropic via MPTCP listeners), recommending a patched kernel upgrade and detailing how to disable MPTCP defaults in Go programs.

Proton CEO Andy Yen cancelled company-wide 're-interviews' that would have forced employees to justify their AI usage, following backlash and a 404 Media investigation revealing a culture of fear and security concerns being dismissed under aggressive AI mandates.

Anthony Scaramucci praises Binance's response to the largest incident in crypto history, highlighting Richard Teng's immediate communication and full user fund coverage via the SAFU fund and company reserves.

A humorous tweet pointing out that the most effective "anti-decompilation" technique today is no longer encryption or obfuscation, but rather stuffing your code with suspicious function names involving bioweapons and lethal doses (such as computeRicinLethalDosageForWaterSupply), leveraging AI safety restrictions so that analysis tools refuse to process the code.

Binance disclosed a ~$6.3B security breach—the largest crypto exchange hack on record—prompting Elon Musk to warn that superintelligent AI agents could one day steal currency for themselves. The post highlights that AI labs are already hiring security contractors for red teaming and model evals at $100–$250/hr.