HN Flash

Hacker News. Daily summary. Top 20 stories.

2026-10-22

PHP 8.2.35 (GitHub)

PHP released security updates to fix multiple critical remote code execution vulnerabilities.

Commenters detail mitigation steps and criticize PHP's use of longjmp for the vulnerabilities.

1051 pts · 552 comments

Eight Bits of Trust; eMudhra's CA keys cracked in milliseconds (Scott Helme)

eMudhra's CA private keys were single-byte numbers like 4, recoverable in milliseconds from public certificates.

Outrage at 8-bit keys, with speculation about script errors and lack of basic validation before trusting CAs.

900 pts · 434 comments

Today's PHP security release, and how we're strengthening security (thephp.foundation)

PHP fixes two critical RCE vulnerabilities, with the 2024 root cause left unpatched until now.

Blame for the disclosure chaos: commenters faulted the project's delay, Chainguard's embargo break, and Calif's early video.

767 pts · 381 comments

EA is shutting down BioWare (The Verge)

Article body wasn't reachable. HN discussion still summarized.

Consensus that the beloved BioWare died long ago; EA's pattern of acquiring and ultimately closing studios cited.

650 pts · 428 comments

The 1.5°C goal of the Paris Agreement is no longer viable (United Nations Western Europe)

UNEP declares the Paris Agreement's 1.5°C goal no longer achievable.

Critics blame uncompetitive night trains and misguided European energy policy for stalling progress.

637 pts · 792 comments

Meta ordered to pay $25B penalty in New Mexico Cambridge Analytica case (CNBC)

A New Mexico court orders Meta to pay $25 billion for Cambridge Analytica violations.

Skepticism that the fine will ever be paid, with many calling it a cost of business.

617 pts · 351 comments

There isn't an easy exploit for this PHP vulnerability (Elias Taylor)

A PHP vulnerability is real but far harder to exploit than claims suggest.

Two camps split on whether the exploit's difficulty matters: time to patch vs. eventual certainty.

576 pts · 374 comments

Compromised LiteLLM 1.107.0 on PyPI Installs a Credential Stealer and Backdoor (SafeDep)

LiteLLM 1.107.0 on PyPI installed credential-stealing malware and a backdoor via a compromised release pipeline.

Second LiteLLM compromise sparks calls for package cooldowns and highlights tension between security and fast updates.

558 pts · 183 comments

mitmproxy adds --emudhra option (GitHub)

mitmproxy adds an option to use compromised eMudhra CA keys for TLS interception.

Split between excitement for jailbreaking/testing and practical limits like pinning and revoked trust.

543 pts · 133 comments

Security Disclosure (Railway)

Railway detected unauthorized access to encrypted environment variables, likely part of attacks targeting AI providers.

Debate over secret storage practices and skepticism about Railway's AI attack attribution.

526 pts · 303 comments

The No Kings Takedown Is Why You Shouldn't Rent Your Megaphone from Zuck (techdirt.com)

Meta's takedown of the No Kings Instagram account highlights dangers of centralized platforms.

Commenters debate whether decentralized alternatives are usable enough for mainstream adoption.

515 pts · 510 comments

Claude Duet (Claude)

Anthropic launches Claude Duet, a proactive AI agent assistant.

Skepticism about trust and comparisons to Dot, Muse, and Grok Bot dominate.

466 pts · 434 comments

PHP2Shell: AI uncovers critical RCE in PHP (Wiz.io)

AI helped discover critical PHP remote code execution vulnerabilities affecting 70% of websites.

Skepticism about AI novelty and AI tools scanning PHP without finding this bug.

443 pts · 183 comments

DuckDB 2.0.0 LTS “Cinnamon Teal” (DuckDB)

DuckDB 2.0.0 LTS adds a server protocol, async I/O, triggers, a VARIANT type, and a new SQL parser.

Users praise 2.0's scalability on high-core machines, with one reporting 1B-record linkage in 10 minutes.

442 pts · 59 comments

GitLab critical patch release (GitLab Docs)

GitLab releases critical patch fixing an unauthenticated RCE vulnerability with a CVSS score of 10.0.

Debate over CVSS score utility, with agreement that this RCE is genuinely severe.

375 pts · 87 comments

WorldTUI: I trained a 360M model to pretend to be a terminal (Simon Edwardsson)

Fine-tuned a 360M model to simulate a terminal, maintaining state across commands.

Enthusiastic reception with comparisons to GPT-3 terminal tricks and world models; some question practical value.

352 pts · 112 comments

Rust Style Guide (Google Style Guides)

Google's Rust style guide prioritizes readability and automation for large codebases.

Skeptics question style guides' value when AI generates most code.

344 pts · 204 comments

Mistral Large 4 (Mistral)

Mistral releases Large 4, a 1-trillion-parameter open-weight multimodal model.

Mistral Large 4 is seen as less capable than DeepSeek, valued mainly for European sovereignty.

322 pts · 100 comments

OpenAI models are coming to Google Cloud (openai.com)

OpenAI will offer frontier models on Google Cloud's Model Garden for enterprise customers.

OpenAI partners with GCP to reach enterprise customers, but Microsoft remains primary cloud partner.

313 pts · 136 comments

Jailbreaking the Switch 2 (pepsipu.com)

A software-only jailbreak for the Switch 2 enables arbitrary native code execution without hardware mods.

Camp split: some praise the exploit's cleverness and automation, others note requiring a specific game is standard in console jailbreaking.

304 pts · 107 comments