security

Discussion of Canonical's response times to CVE vulnerability disclosures, highlighting concerns about patching speed for Ubuntu and related packages.

Commentary on the risk that AI agents can modify their own budget limits, arguing enforcement and storage of spending caps must live outside anything the agent can write to. The author shares how their runtime, mainsheet, enforces cost caps as a code-level gate at the tool-call boundary rather than a prompt rule.

Firefox 的 2026 年第三季度安全与隐私通讯，涵盖更快的发布周期、TLS 配置透明度警告、Mozilla Root Store Policy v3.1、mailto 链接防骚扰以及扩展本地文件访问权限限制等安全与隐私改进。

China Medical University Hospital, Taiwan's second-largest healthcare provider, uses VMware Cloud Foundation and vDefend to deploy compliant IT in under half a day while protecting clinical workloads with Zero Trust security.

A user publicly accuses Railway employee Jake Runzer of attempting to breach their infrastructure, sparking a security-related allegation against the hosting platform.

A post argues that coding agents often claim to fix security bugs while leaving the original PoC still exploitable, and proposes freezing the PoC, evidence, and missed check as fixtures for every rejected action so patches are verified against the trigger, not just the diff.

OpenAI is more than quadrupling bug bounty reward ranges for medium, high, and critical vulnerabilities, adding quality multipliers, a tiered loyalty program, priority campaign bonuses, and retroactive payments for recent awards.

GitHub Next introduces enclaves, an experimental security primitive for GitHub Agentic Workflows that lets agents work across repositories while deterministically constraining what they can communicate back, mitigating private-to-public data leaks like the GitLost attack.

Google revealed that one of its Android manufacturing partners, Shenzhen-based Videostrong Technology, secretly built a backdoor into 1.3 million set-top boxes and other Android devices, turning them into a botnet used for ad fraud and malware distribution. Google has dismantled the backdoor, cut ties with the manufacturer, and updated its device review process.

The author compares agent guardrail approaches deployed at different layers this week — traffic proxies, browser secret-scrubbing, cryptographic agent identity for MCP, and GitHub Actions audit logs — and argues tool-call gating in the harness catches irreversible actions that wire-level proxies miss. Sharing their own Claude Agent SDK runtime 'Mainsheet' with a signed-receipt hash chain, they ask practitioners where enforcement should live and what each layer catches that the other misses.

Binance says it has fully allocated funds to cover its $3.3B hack losses, with 95% recovered and the remaining 5% in less liquid tokens still being converted, as top-10 withdrawals are restored.

A blogger reverse engineers a 2014 Renault Latitude's TomTom Carminat LIVE head unit to restore the missing clock-settings menu, discovering the menu definition is only loaded from a navigation SD card. The writeup covers extracting the ext3 root filesystem from firmware updates and patching the Linux-based device.

Tor Browser 16.0 is released for Desktop and Android, based on Firefox ESR 153, featuring a redesigned Settings experience, a new generic window titles option to prevent snooping, and an audit of over 270 Firefox changes affecting privacy and security.

A proposed Rust macro technique that makes reverse engineering attacks over 700x more expensive while only adding small linear overhead to the defender, with suggestions it could be made exponentially costly in constant time similar to RSA.

Research on recognizing keystrokes using WiFi devices, raising both sensing capability and privacy/security implications for wireless monitoring.

A breakdown of six options for safely running code generated by AI agents, ordered from least to most isolation: local Docker containers, E2B, Modal Sandboxes, Daytona, Vercel Sandbox, and self-hosted Firecracker or gVisor, with advice to block outbound network and set resource limits by default.

TAWA GUARD introduces a pre-model input check for AI agents that screens tool output, documents, and email for hidden instructions, returning allow, flag, or block decisions with user-configurable thresholds.

DeepSeek's agent team published a design document arguing the bottleneck for AI agents is human attention rather than model capability, proposing a Router/Worker session architecture where a person can realistically track only 3-5 active agent sessions.

The NOIR tool on GitHub can scan domestic phones via adb over USB, generating a list of domains that exfiltrate data such as location and contacts, along with rollback commands, for checking a phone's privacy leaks.

This paper introduces Aligned RAG Evaluation, a matched-query protocol showing that attack success rate (ASR) hides defense-induced failures in RAG knowledge-poisoning defenses, distinguishing prevented attacks from newly introduced target-positive answers on HotpotQA and Natural Questions.