惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

U
Unit 42
S
Security Affairs
Engineering at Meta
Engineering at Meta
MongoDB | Blog
MongoDB | Blog
Microsoft Security Blog
Microsoft Security Blog
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
美团技术团队
月光博客
月光博客
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
有赞技术团队
有赞技术团队
博客园 - 叶小钗
Hugging Face - Blog
Hugging Face - Blog
Microsoft Azure Blog
Microsoft Azure Blog
V
V2EX
Vercel News
Vercel News
阮一峰的网络日志
阮一峰的网络日志
腾讯CDC
C
Cisco Blogs
T
Tor Project blog
The Hacker News
The Hacker News
雷峰网
雷峰网
MyScale Blog
MyScale Blog
博客园 - 司徒正美
AWS News Blog
AWS News Blog
GbyAI
GbyAI
Y
Y Combinator Blog
D
DataBreaches.Net
Simon Willison's Weblog
Simon Willison's Weblog
S
Securelist
The GitHub Blog
The GitHub Blog
S
SegmentFault 最新的问题
T
Tenable Blog
L
LangChain Blog
M
MIT News - Artificial intelligence
N
Netflix TechBlog - Medium
The Cloudflare Blog
A
About on SuperTechFans
IT之家
IT之家
F
Fortinet All Blogs
Security Latest
Security Latest
Threat Intelligence Blog | Flashpoint
Threat Intelligence Blog | Flashpoint
NISL@THU
NISL@THU
爱范儿
爱范儿
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
cs.AI updates on arXiv.org
cs.AI updates on arXiv.org
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
W
WeLiveSecurity
A
Arctic Wolf
I
Intezer
Application and Cybersecurity Blog
Application and Cybersecurity Blog

Hacker News - Newest: "AI"

AI can't read an investor deck AI as an attorney? Student uses ChatGPT, Gemini to sue UW over alleged racial discrimination Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval GitHub - MeepCastana/KubeezCut: Free Web based video editor GitHub - GenAI-Gurus/awesome-eu-ai-act: Curated tools, official sources, OSS, templates, and guides for EU AI Act compliance. Can AI judge journalism? A Thiel-backed startup says yes, even if it risks chilling whistleblowers Coming soon: 10 Things That Matter in AI Right Now DARPA built an AI to fact-check enemy weapons claims What explains heterogeneity in AI adoption? When AI Meets Muscle: Context-Aware Electrical Stimulation Promises a New Way to Guide Human Movements - Department of Computer Science AI Changed How We Build. It Did Not Change What Matters. Linux rules on using AI-generated code - Copilot is OK, but humans must take 'full responsibility for the… Meta spins up AI version of Mark Zuckerberg to engage with employees Code Mode: Let Your AI Write Programs, Not Just Call Tools | TanStack Blog GitHub - Delavalom/graft: Go framework for building AI agents. Type-safe tools, multi-provider (OpenAI, Anthropic, Gemini, Bedrock), zero vendor SDKs. India's TCS tops estimates, says new AI models did not dent services demand Gen Z's fading AI hype Strong feeling: we are in a folded AI reality GitHub - machinarii/total-recall-catalog: A reference catalog of latest knowledge retrieval, memory & RAG systems GitHub - mensfeld/code-on-incus: Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops threats automatically.. Quantization, LoRA, and the 8% Problem: Benchmarking Local LLMs for Production AI Iran war: We spoke to the man making Lego-style AI videos that experts say are powerful propaganda Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks GitHub - immartian/bellamem: Persistent belief-graph memory for AI agents. Retrieves decisive context by importance — not recency, not RAG, not /compact. recursive-mode: The Repo-Native Operating System for AI Engineering After the attack on Sam Altman's home, will AI CEO's go on the offensive? The biggest advance in AI since the LLM Opus 4.6 vs GPT 5.4 One Prompt Unity World Generation Test “AI polls” are fake polls Client Challenge Can AI be a 'child of God'? Inside Anthropic's meeting with Christian leaders How to Switch AI Chatbots and Why You Might Want To GitHub - MattMessinger1/agentic_refund_guardrail: Safe refund policy layer for AI agents — Python + TypeScript. Same behavior, shared tests. Adam/papers/emergent_values_whitepaper.md at master · strangeadvancedmarketing/Adam Ask HN: How do you stop playing 20 questions with your AI coding tools How far can automation and AI support psychotherapy? - @theU GitHub - stagas/rtdiff: realtime git diff gui and AI-assisted commits A Mac Studio for Local AI — 6 Months Later A History of the Early Years of AI at the University of Edinburgh Why AI Coding Tools Still Feel Stuck on Localhost MSN AI Datacenters Are Becoming Strategic Targets twitter.com Penn Researchers Use AI to Surface Unreported GLP-1 Side Effects in Reddit Posts Show HN: MoodSense AI (ML and FastAPI and Gradio, Deployed on Hugging Face) Moodsense Ai - a Hugging Face Space by aman179102 AI models are terrible at betting on soccer—especially xAI Grok GitHub - xialeistudio/echoic GitHub - HimashaHerath/github-dev-wrapped: AI-powered weekly GitHub activity reports deployed to GitHub Pages GitHub - alejandrobalderas/claude-code-from-source: Architecture, patterns & internals of Anthropic's AI coding agent — reverse-engineered from source maps AI and Tech brief: Ireland ascendant GitHub - Titovilal/context0: Context0 - Never Surrender Training for a Marathon with an AI Coach: What Worked and What Didn't Cyber Pulse: Agentic Intel - Apps on Google Play I Built an AI PR Reviewer That Catches Bugs by Not Looking for Bugs Gen Z workers are so fearful AI will take their job they’re intentionally sabotaging their company’s AI rollout | Fortune How AI Is Reimagining the Game of Golf–For Both Players and Courses GitHub - nattergabriel/reseed: A CLI tool for managing and distributing agent skills across projects Is SVG the final frontier? My AI workflow evolved from prompts to a near-autonomous workflow MLSharp Help - 3DGS Viewer & Generator I put my cognitive field based AI's runtime on GitHub Is Numble the first AI-proof game? A3: Kubernetes for autonomous AI agent fleets | Emergent Principles Deepali Vyas ("The Elite Recruiter") GitHub - msmarkgu/RelayFreeLLM: A restful API designed to route user prompts to various AI model providers. Unionized ProPublica staff are on strike over AI, layoffs, and wages Unleashing the Advantage of Quantum AI We're heading for an AI-fueled 'dementia crisis,' brain scientist warns The AI-Assisted Breach of Mexico's Government Infrastructure [pdf] GitHub - stef41/lmscan: 🔍 Detect AI-generated text and fingerprint which LLM wrote it. Open-source GPTZero alternative. Zero dependencies, works offline. MSN GitHub - visionscaper/collabmem: Enabling long-term collaboration with Agentic AI - building up episodic and world model memory over time with in-context awareness We gave an AI a 3 year retail lease in SF and asked it to make a profit | Andon Labs AI Code is Hollowing Out Open Source, and Maintainers are Looking the Other Way What leaked "SteamGPT" files could mean for the PC gaming platform's use of AI AI is the boss at this retail store. What could go wrong? GitHub - Wuzu11517/agentic-proxy: Local proxy meant to help reduce With Drones, Geophysics and ArtificiaI Intelligence, Researchers Prepare to Do Battle Against Land Mines A Single Operator, Two AI Platforms, Nine Government Agencies: The Full Technical Report 在 Steam 上购买 FriedrichAI: Offline AI 立省 10% GitHub - inevolin/resume-cli: Hit Claude usage limits? Resume any AI coding session elsewhere. Switch tools at zero friction. GitHub - atripati/ark: AI Runtime Kernel — a context operating system for AI agents. Eliminates tool bloat, loads only what’s needed, and gives LLMs their reasoning space back. How to Build a Secure AI PR Reviewer with Claude, GitHub Actions, and JavaScript This Startup Wants You to Pay Up to Talk With AI Versions of Human Experts Intel Arc Pro B70 Brings 32GB VRAM to Local AI for $949 WordPress 7.0: The Good, the AI, and the Still Missing AI on the couch: Anthropic gives Claude 20 hours of psychiatry IatroBench: Pre-Registered Evidence of Iatrogenic Harm from AI Safety Measures AI Agents Know About Supabase. They Don't Always Use It Right. The history and future of AI at Google, with Sundar Pichai Inside an AI‑enabled device code phishing campaign How Meta Used AI to Map Tribal Knowledge in Large-Scale Data Pipelines AI for Systems: Using LLMs to Optimize Database Query Execution Forecasting the Economic Effects of AI Introducing Tinker: Play with AI, bring your ideas to life AI sheds light on an ancient gaming mystery People really hate AI but not as much as Iran—or Democrats | Fortune What is an AI Product Engineer? Phoebe Gates wants her $185 million AI startup to succeed with 'no ties to my privilege or my last name': 'I have a chip on my shoulder' | Fortune
Open Source is the Tip of the Iceberg: Why Proprietary Software, Hardware and Protocols Face Greater AI-Driven Security Risk
dwitcher · 2026-05-06 · via Hacker News - Newest: "AI"

When Anthropic launched Project Glasswing in April 2026 – a consortium of eleven major companies deploying its Claude Mythos Preview model to find vulnerabilities in critical open-source software − the cybersecurity industry applauded. And rightfully so: Mythos discovered bugs that had hidden in heavily-audited codebases for decades.

But Glasswing’s focus on open source, while valuable, addresses the most visible part of the problem. Open-source software has always benefited from community review, etc. The software that nobody has been looking at – proprietary binaries, embedded firmware, legacy protocols, even chip microcode – carries a far larger and more dangerous accumulation of undiscovered vulnerabilities. And the same AI capabilities that make Glasswing possible are about to expose all of it.

Code that nobody looks at accumulates hidden bugs. Proprietary software operates under a fundamentally different model. Its security posture has historically relied on a simple premise: if attackers can’t read the source code, they’ll have a harder time finding bugs. This isn’t better security – it’s security by obscurity.

The Binary Barrier Is Falling

The traditional answer was it doesn’t matter because attackers can’t read proprietary source code. They only get compiled binaries – stripped of variable names, comments, and structure. That answer is becoming obsolete.

What wasn’t solved was the human bottleneck. A typical security audit covers only a fraction of a codebase – auditors triage by intuition, focus on high-risk surfaces, and leave vast swaths of code untouched. LLMs eliminate this bottleneck. Claude Mythos Preview can take a closed-source, stripped binary, reconstruct plausible source code, and systematically analyze it for vulnerabilities.

Evidence Is Already Here: Edge Devices Under Siege

This isn’t a theoretical risk. That exposure is already being collected – violently – on one category of proprietary software: network edge devices. Firewalls, VPN gateways, load balancers, and secure access appliances have experienced an unprecedented wave of critical zero-day discoveries. According to Verizon’s 2025 DBIR, exploitation of edge device vulnerabilities increased eight-fold in a single year. The median time from vulnerability disclosure to active exploitation is now zero days. The median time to patch: 30 days.

These devices were the poster children for security-through-obscurity: proprietary firmware, closed-source code, no ability to install endpoint detection agents, and internet-facing by design. They were considered secure in part because their code was hard to analyze. That assumption has collapsed. Perhaps most alarming: over 40% of exploited vulnerabilities in 2025 involved end-of-life products –devices that will never receive a patch.

The Long Tail: Where Proprietary Software Hides

The edge device crisis is a preview.

  • Hospital infusion pumps, MRI machines, and patient monitors run proprietary firmware that may not have been updated since the device was certified – sometimes a decade or more.
  • SCADA controllers and PLCs managing power grids, water treatment plants, and manufacturing lines run proprietary firmware implementing protocols designed in the 1980s.
  • A modern vehicle contains over 100 million lines of code distributed across dozens of electronic control units (ECUs), sourced from a fragmented supply chain of tier-one and tier-two suppliers.
  • Large organizations run SAP, Oracle, and custom enterprise applications with modules that haven’t been substantially rewritten in 15-20 years. Their internal codebases rarely see external security review.

Beyond Software: Protocols That Can’t Be Patched

Software vulnerabilities, however dangerous, can at least be patched. Protocol vulnerabilities are a different problem entirely – because the flaws are in the specification itself, not in any particular implementation. Fixing them means replacing the protocol.

  • SS7 (Signaling System 7), designed in the 1970s for a closed network of trusted telecom operators, carries no authentication whatsoever.
  • BGP, the protocol that routes all internet traffic, allows any network to announce any route with zero verification.
  • Industrial protocols – Modbus, DNP3, BACnet – were designed for serial links between trusted devices in isolated environments. Secure variants exist on paper (Modbus over SNMPv3 per RFC 8502; DNP3 Secure Authentication per IEC 62351), but real-world adoption is negligible.

AI changes that calculus. An LLM that can read protocol specifications, cross-reference network scan data, and understand deployment topology does not need to discover new protocol weaknesses. It needs to operationalize known ones against specific targets at scale. The economics shift from “one attacker, one target” to “one AI system, thousands of targets”

Beyond Software: Chips and Microcode

The layer beneath software presents an even more daunting challenge. And here, the common assumption – that chip-level vulnerabilities require access to chip design files — is demonstrably wrong. Every major CPU vulnerability discovered in the past decade was found without access to the manufacturer’s hardware design.

  • Spectre and Meltdown were discovered through timing measurements and deep reasoning about speculative execution, building on years of academic cache side-channel research and working from public architecture manuals.
  • Google’s Reptar was found through targeted instruction fuzzing — testing how CPUs handle unusual instruction sequences and redundant prefixes.
  • Downfall was found by methodically testing memory-accessing Gather instructions for transient execution data leakages.

These approaches – reasoning about architectural documentation, behavioral observation, intelligent fuzzing – are precisely the tasks that LLMs accelerate. The critical difference from software: you cannot simply patch a chip. Microcode updates are partial mitigations that often carry performance penalties. Silicon-level flaws cannot be fixed without a new chip revision.

The Force Multiplier: Cross-Layer Exploit Chaining

Each layer of vulnerability – software, protocols, hardware – is concerning on its own. The compounding danger emerges when AI systems begin chaining vulnerabilities across layers. The next frontier is chains that span layers entirely:

  • Protocol implementation bugs that trigger a microarchitectural side channel.
  • Firmware vulnerabilities that expose a cryptographic key used by a flawed protocol.
  • Browser exploits that leverage a CPU speculation bug to escape a sandbox.

Human exploit developers rarely achieve this because cross-domain expertise is rare.

What Needs to Change

Project Glasswing is a good start. But it addresses the most visible and already best-defended category of software. The industry’s response needs to be broader:

  • Assume obscurity provides zero protection. Any organization shipping proprietary binaries, embedded firmware, or custom protocol implementations should operate under the assumption that AI-powered analysis will find their vulnerabilities — if it hasn’t already. Security-through-obscurity is now a failed strategy.
  • Extend AI-powered auditing into the long tail. Glasswing already partners with major vendors – AWS, Microsoft, Cisco, Broadcom, NVIDIA, JPMorgan – to scan proprietary codebases, which proves the approach works. But these are the head of the distribution: well-resourced companies with security budgets and strong incentives to engage. The real exposure sits in the long tail — thousands of organizations whose code has never been independently reviewed, whose installed bases run for decades, and who have no existing relationships with AI security firms.
  • Prioritize the systems that can’t be patched quickly. Software can be updated in hours. Firmware updates take weeks to months and protocol changes take years. Silicon can’t be fixed at all. Defensive investment should be weighted toward the layers where remediation is slowest, because those are the layers where discovered vulnerabilities persist longest. With over 40% of exploited vulnerabilities targeting end-of-life products that will never be patched, the installed base of unpatchable systems is itself an attack surface.
  • Prepare for cross-layer attacks. Security teams organized in silos – network security, application security, and hardware security – will miss the attack chains that cross boundaries. Red teams need to develop cross-domain thinking, or more practically, deploy AI systems that already think cross-domain.
  • Close the response gap for edge devices. The industry-wide picture is uneven – for most CVEs, defenders still have days or weeks before exploitation becomes widespread. But for critical vulnerabilities in internet-facing edge devices, the median time from disclosure to exploitation is now zero days while median time to patch is 30. Organizations that cannot patch those devices within hours, not weeks, need compensating controls that assume the devices are compromised.

The era in which hidden code meant hidden bugs is ending. The edge device crisis of 2024-2026 is the early tremor. Project Glasswing illuminates one corner of the landscape. The rest of the iceberg – proprietary firmware, legacy protocols, chip microcode, and the cross-layer chains that connect them – is still underwater, and it is considerably larger than what’s visible above the surface.