Top 10 Hacker News posts, summarized
HN discussion
(708 points, 653 comments)
Belgium has announced it will stop decommissioning its nuclear power plants, with Prime Minister Bart De Wever stating the government will pursue nationalization of the facilities. The decision reverses a 2003 phase-out policy and aims to create a safer, more affordable, and sustainable energy supply with reduced reliance on fossil fuel imports. Belgium, which has seven nuclear reactors (three of which are already offline), is currently in exclusive negotiations with operator ENGIE to acquire its complete nuclear fleet, including assets and liabilities, with a basic agreement expected by October.
The Hacker News discussion focused on several key points. Many commenters argued for maintaining existing nuclear plants as a stable, clean energy source but were skeptical of building new ones due to high costs and timeframes. Critics raised concerns about Belgium's ability to manage the plants effectively given its financial struggles and poor track record with semi-public companies. Other commenters highlighted the strategic importance of nuclear energy for energy independence, while some questioned the feasibility of restarting reactors and the financial implications. The debate also touched upon broader EU energy policies, the challenges of nuclear waste storage, and political motivations behind the policy shift.
HN discussion
(852 points, 488 comments)
The article reports that Anthropic's Claude Code service is refusing user requests or imposing extra charges when commits mention "OpenClaw," an open-source alternative to Claude. Users have found that even references to the term in commit messages or project files trigger these automated responses, with some experiencing immediate disconnections and hitting usage limits. This behavior has raised concerns about the implementation and transparency of Anthropic's content moderation and pricing policies.
The top HN comments criticize Anthropic's approach as poorly implemented, likely relying on simplistic regex matching that leads to false positives. Users express frustration over the opaque, seemingly arbitrary enforcement, arguing it undermines trust and highlights potential resource constraints at the company. The discussion also includes calls for more transparent, usage-based pricing and suggestions for better technical solutions, while others note the incident as a reason to cancel subscriptions and switch to local or alternative models.
HN discussion
(562 points, 205 comments)
Mozilla has formally opposed Chrome's Prompt API, a proposal for a web-based interface to interact with language models. The opposition, articulated by Jake Archibald of Mozilla, centers on two main concerns: the risk of creating model-specific prompts that would lock developers into a single implementation, and the lack of model neutrality in the terms of service. These issues could force other browser vendors to license Google's model or create a compatible version, effectively cementing Google's dominance. The article notes that Chrome has already published an intent-to-prototype for the API.
The Hacker News discussion is dominated by skepticism of Google's motives, with many commenters accusing the company of abusing its dominant market position to push a proprietary standard. The conversation highlights a broader frustration with the "monoculture" of the web and a desire for more open, standardized solutions, such as public domain models that would be consistent across browsers. Some commenters draw parallels to past Google initiatives like FLoC and the Topics API, which were also seen as anti-competitive. Others question the necessity of such an API, suggesting it could be better handled by web developers themselves on a case-by-case basis, while a few worry about privacy, performance, and the precedent of building content restrictions into browser APIs.
HN discussion
(307 points, 237 comments)
The article discusses CVE-2026-31431 ("CopyFail"), a critical Linux kernel vulnerability introduced in kernel 4.14 and fixed in versions 6.18.22, 6.19.12, and 7.0. Affected older LTS kernels (6.12, 6.6, 5.15, etc.) remain unpatched due to complex backporting. Gentoo developer Sam James shared a workaround patch and criticized the lack of communication between the kernel security team and distributions, noting that vulnerability reporters must manually notify downstream distros via the linux-distros ML for heads-ups. The vulnerability allows local privilege escalation, and public disclosure occurred before distribution patches were widely deployed.
Top HN comments criticized the Linux kernel's security disclosure process, emphasizing that the burden of notifying distributions should not fall on vulnerability reporters. Users like semiquaver and whatevaa argued that the kernel team, not individual reporters, is responsible for coordinating with downstream consumers. Others highlighted practical impacts, such as shared hosting risks (xeeeeeeeeeeenu) and workarounds (e.g., GranPC's eBPF-based mitigation). System-specific impacts were noted, with seniorThrowaway confirming Ubuntu patches and lrvick stating stagex was unnotified. KingMachiavelli advocated for security hardening (e.g., default `nosuid` mounts), while swinglock clarified NixOS was still affected despite mitigations.
HN discussion
(373 points, 162 comments)
Unable to fetch article: HTTP 403
The discussion centers on Spain's parliamentary action against LaLiga's court-ordered IP blockades targeting Cloudflare addresses during football matches to combat piracy. Key insights highlight the disproportionate collateral damage: legitimate websites using shared Cloudflare IPs became inaccessible, rendering the blocks ineffective against piracy while severely disrupting unrelated services. Reactions emphasize widespread condemnation of the approach as absurd, counterproductive, and technologically flawed, with multiple users noting the irony that Cloudflare's own error pages were blocked. Critics argued the enforcement lacked proportionality, ignored a clear "stopping principle," and prioritized LaLiga's commercial interests over internet functionality, calling the policy a "stupid prize" for irresponsible legal tactics. There was also skepticism about Cloudflare's technical capability to implement granular, time-based blocking, and concern that ISPs like Telefónica suffered further reliability issues.
HN discussion
(374 points, 105 comments)
Unable to fetch article: HTTP 403
The Hacker News discussion centers on the historical significance of Room 641A, a former AT&T facility used for NSA surveillance, highlighting its role in early 2000s mass surveillance programs. Commenters note this is "old news" contemporaneous with Snowden revelations, emphasizing its relevance to current debates over Section 702 FISA reauthorization and Senator Wyden's calls for declassified public debate on surveillance authorities. Some draw connections to legal loopholes like the Five Eyes alliance for domestic spying, while others reference technical aspects such as Perfect Forward Secrecy's impact on surveillance efficacy.
The thread also includes personal accounts of alleged government surveillance targeting, like one user describing infiltration and harassment after publicly criticizing the CIA, alongside broader critiques of the surveillance-industrial complex. Political polarization is noted as a factor stifling reform, with bipartisan agreement on surveillance rendering it a "non-issue," while concerns persist about the erosion of privacy norms and the legal justifications for domestic surveillance. The discussion underscores the tension between national security imperatives and civil liberties, with some questioning the sustainability of democratic governance under advanced technological surveillance.
HN discussion
(269 points, 166 comments)
IBM released Granite 4.1, a family of Apache 2.0 licensed dense transformer models in 3B, 8B, and 30B sizes. The key highlight is the 8B model, which matches or outperforms the previous-generation 32B MoE model across benchmarks like ArenaHard, BFCL V3, and GSM8K. This performance is attributed to a 15-token training process with five distinct phases focusing on high-quality data and a sophisticated filtering system for instruction-fine-tuning. The models are designed for enterprise predictability, avoiding complex architectures like MoE and achieving a 512K context window through staged training and model merging, while also maintaining performance at shorter context lengths.
The HN discussion focused on the practical performance and trade-offs of the Granite 4.1 models. Users noted that the 8B size is a strong contender for local and cost-sensitive use cases, though some pointed out it is outperformed by other models like Qwen 3.5 on certain benchmarks. There was significant debate about IBM's decision to move away from MoE architectures, with some attributing it to a focus on training efficiency and inference predictability. Commenters also highlighted the potential of the smaller 3B model for edge deployment and expressed a desire for an accompanying embedding model. The article's authorship was also criticized, with many commenters dismissing it due to its perceived robotic writing style.
HN discussion
(309 points, 118 comments)
Rivian has introduced a feature that allows users to disable all internet connectivity in their electric vehicles. This option, which is a toggle in the settings for Canadian customers and requires a service appointment for those in the U.S., aligns with the company's mission to preserve nature and improve its products. Disabling the connectivity restricts certain functions, such as over-the-air updates, navigation, and lane-keeping assistance.
Hacker News users praised Rivian for offering the privacy feature, with some comparing it to other software settings like "disable_ai" and applauding the company for listening to customer feedback. However, several concerns were raised, including the inconvenience of needing a service appointment in the U.S. versus the simple toggle in Canada, the potential loss of critical safety features like lane-keeping assistance, and the lack of a physical disconnect for antennas. Commenters also debated the implications for safety recalls and criticized other automakers for similar practices.
HN discussion
(285 points, 110 comments)
LinkedIn scans visitors' browsers for over 6,000 Chrome extensions on every visit, according to a detailed technical analysis. The company probes for specific files within each extension to build a detailed software inventory, which is then encrypted and injected as an HTTP header into all subsequent API requests during a user's session. This practice, which dates back to at least 2017 and is not disclosed in LinkedIn's privacy policy, allows the company to link browsing habits and installed software to a user's verified professional identity. The author notes that this is not standalone fingerprinting but part of a larger system called APFC (Anti-fraud Platform Features Collection), and that a criminal investigation is now open in Germany related to this activity.
The HN discussion expresses strong condemnation of LinkedIn's actions, with one commenter stating, "Good. These companies deserve each and every stone thrown at them, and much more." There is significant debate about the purpose of the scans, with many commenters believing it is for surveillance and to identify job seekers, while others argue it is likely an effort to combat scraping and fraud. Some users are questioning their browser choice, with one asking if Safari or Firefox would prevent this behavior. Another commenter critically analyzed the article's claims, arguing it contains "ragebait journalism" and that the scanned extensions are primarily data extraction tools and scrapers, not benign applications like those for neurodivergent users as the article implies.
HN discussion
(296 points, 95 comments)
The PyTorch Lightning AI library (package 'lightning') was compromised in versions 2.6.2 and 2.6.3 (published April 30, 2026), containing a hidden JavaScript payload that executes upon import. The malware steals credentials, tokens, environment variables, and cloud secrets while propagating via npm to publish infected droppers. It targets developers by injecting persistence hooks into Claude Code and VS Code configurations and pushes malicious GitHub Actions to exfiltrate data. The attack is attributed to the "Mini Shai-Hulud" threat actor, using Dune-themed nomenclature and multi-channel data exfiltration (HTTPS POST, GitHub dead-drops, public repos, and victim repo pushes). Affected environments (local/CI/cloud) must be audited for compromised files and rotated credentials.
Hacker News comments highlighted concerns over supply chain vulnerabilities and ecosystem dependencies. Key reactions included references to the Dune theme ("Bless the Maker"), observations of widespread propagation (2.2K GitHub repos with attacker descriptions), and debates about dependency management practices like pinning and lockfiles. Users questioned the compromise mechanism (e.g., whether it originated from a malicious PR or mirror) and criticized the potential delay in warnings from the maintainers. Discussions also touched on broader trends of increasing supply chain attacks, the risks of unchecked dependency sprawl in ML/Python ecosystems, and calls for improved security tooling and developer awareness.
Generated with hn-summaries