Latest

Latest News

View All News

10th Gen Honda Civic Updates Are Signed with AOSP Test Keys

Honda's 10th-gen Civic headunit accepts signed AOSP update files during USB-based firmware updates, but the vendor left the publicly-known AOSP test key in `res/keys`. Because the modified recovery binary retains stock AOSP `verify_file` signature logic, any properly formatted USB drive signed with that test key will be accepted. This enables arbitrary code execution on the headunit via the update path with only physical access to the front USB port—an "evil valet" attack applicable to any journalist or driver whose car is temporarily accessible to an attacker.

Jun 14, 2026juniperspring.org

Running DOS on Behringers DDX3216 with a DIY x86-Bios from Scratch

The Behringer DDX3216 digital mixer uses an AMD Elan SC300 386 SoC, enabling bare-metal x86 software execution. By reverse-engineering the hardware and writing a custom BIOS from scratch,

Jun 14, 2026chrisdevblog.com

Google proposes Open Knowledge Format based on Markdown

Google Cloud announced the Open Knowledge Format (OKF), an open specification formalizing the LLM-wiki pattern into a portable, interoperable, vendor-neutral standard. OKF v0 provides a Markdown-based structure for representing metadata, context, and curated knowledge needed by AI systems, particularly agentic systems requiring relevant context beyond what foundation models inherently possess. This aims to improve data sharing and enable more accurate, actionable outputs by supplying the structured context modern AI needs.

Jun 14, 2026cloud.google.com

Introduction to the experience of rendering Arabic typography&its technical debt

An article details systemic technical debt in Arabic typography rendering, illustrated through real-world bugs: mixed-content Arabic prose fails to justify, legacy PDF libraries lack shaping engines, and a search index breaks due to inconsistent Unicode codepoint versions (1991 vs. 1995). These issues stem from foundational neglect of bidirectional text and Arabic shaping in software stacks, with each bug often presenting as isolated despite sharing root causes. The author argues that such

Jun 13, 2026lr0.org

Leaving Mozilla

After more than 15 years, jr conlin is leaving Mozilla with his last working day on June 12, 2026, using over 200 hours of accrued vacation before his official departure on July 21. In his farewell, he underscores the critical importance of mentoring, especially in an organization of introverts, as a tool for personal and career growth independent of experience level. He also reminds technical staff that they are part of a mission-driven project—building a

Jun 13, 2026blog.unitedheroes.net

There is a shadow hanging over this Fable thing

The US government directed Anthropic to disable access to its Fable and Mythos models for any foreign national worldwide, including within the US and among Anthropic employees. Citing the directive as impossible to enforce selectively, Anthropic responded by disabling all access to Fable and Mythos. Users encountered sudden model unavailability, such as the error "Sorry! This model doesn’t exist any more!", disrupting agent-based coding workflows and custom

Jun 13, 202612gramsofcarbon.com

Jun 13, 2026

Solid-state batteries are now powering EVs in the real world

Factorial Energy and Stellantis have begun on-road testing of solid-state batteries in a Dodge Charger Daytona development vehicle, the first such integration in North America. The 77Ah FEST cells achieve 375 Wh/kg energy density, maintain over 600 cycles, and support ultra-fast charging from 15% to 90% in 18 minutes. The battery also operates across a wide temperature range (-30°C to 45

Jun 13, 2026

Renault: Electric motors with no rare earths

Renault Group has commercialized electrically excited synchronous motors (EESM) since 2012, a technology that eliminates the need for rare-earth metals by using an electromagnetic field instead of permanent magnets. This contrasts with the dominant industry approach — permanent-magnet synchronous motors — found in approximately 90% of electric vehicles, which require rare earths. Renault’s EESM design achieves high efficiency without rare-earth materials, though the motors are slightly larger than equivalent permanent-magnet units. The company positions this as a long-standing differentiator in the EV powertrain

Jun 13, 2026

How to setup a local coding agent on macOS

The author details a local coding agent setup on macOS using llama.cpp with Metal acceleration, a Gemma 4 26B-A4B GGUF model (Q4_K_XL, ~16 GB), a Q8 MTP draft model for speculative decoding, and Pi as the terminal agent. The configuration also includes the Gemma 4 multimodal projector for image input, targeting an OpenAI-compatible API for tool integration. Tested on an Apple M1 Max with 64

Jun 13, 2026

I Am Not a Reverse Centaur

Miguel Grinberg reports that nearly all unsolicited pull requests on his open source projects are now generated by LLMs, forcing him to spend increasing time reviewing machine-produced code. He rejects the label "reverse centaur"

Jun 13, 2026

"Don't You Just Upload It to ChatGPT?"

A freelance translator recounts being asked if they simply "upload it to ChatGPT," highlighting a common misconception about AI's ability to replace human language professionals. The anecdote underscores that translation involves far more than raw text conversion—requiring cultural nuance, context, and subject-matter expertise that current large language models lack. For technical readers, this reinforces the gap between narrow AI tools and the cognitive flexibility needed for

Jun 13, 2026

The Unsung Hero of the Lord of the Rings (2025)

Tom Bombadil, intentionally left enigmatic by Tolkien, is the only character wholly immune to the Ring’s power and represents the author’s underlying ethics of resisting evil through detachment from dominion. The article argues that Bombadil’s role—often overlooked and omitted from Peter Jackson’s films

Jun 13, 2026

Appeals court upholds FTX co-founder Sam Bankman-Fried's fraud conviction

The U.S. Second Circuit Court of Appeals upheld Sam Bankman-Fried’s conviction on seven counts of fraud and conspiracy related to the collapse of FTX. The ruling affirmed that Bankman-Fried misappropriated billions in customer deposits to fund personal ventures, political donations, and cover losses at Alameda Research. Technical readers should note that the court rejected arguments regarding the trial judge’s evidentiary rulings and jury instructions, which had centered on complex financial instrument valuations and the opacity of FTX’s internal accounting systems. The decision solidifies the legal framework for prosecuting misrepresentations in cryptocurrency

Jun 13, 2026

Ryanair dark UX patterns summer 2026 refresher

Ryanair’s 2026 check-in process requires users to successfully navigate nine distinct stages of dark UX patterns to avoid extra charges, including a deceptive insurance opt-out buried in a country list, a modal pop-up for priority boarding with no “No” button, and a warning about gate fees for selecting “1 Small Bag.” The sequence systematically exploits user fatigue and confirmation bias, with each step designed to maximize incremental revenue through forced choices and hidden defaults. This update refreshes classic dark patterns—such as random seat selection

Technology

Latest Technology

All Technology

Phoenix LiveView 1.2

Phoenix LiveView 1.2.0 introduces colocated CSS within HEEx templates, leveraging a `:type` attribute (e.g., `:type={MyApp.ColocatedCSS}`) to extract and process styles at compile time alongside the existing colocated JavaScript pipeline. The release enables automatic scoping of colocated styles via the CSS `@scope` rule, with LiveView generating unique root attributes in the rendered HTML to isolate component styles. Developers can adopt the update by changing the `phoenix_live_view` dependency to `~>

Jun 15, 2026phoenixframework.org

No, everyone is not using AI for everything

Contrary to popular narratives, most people are not regular AI users. Gallup year-over-year data shows Gen Z AI adoption has stalled, with only ~31% using it monthly and 19% never using it. Microsoft's telemetry-based "AI Diffusion" site reports just over 30% of the US working-age population engages with AI, a figure that increased only 3 percentage points since late 2025. The assumption that “everyone is using AI for everything” is not supported by adoption and usage data.

Jun 15, 2026gabrielweinberg.com

How to Earn a Billion Dollars

YC has funded about 6,500 companies since 2005, producing roughly 30 billionaires primarily through high-growth startups. The essay refutes a politician’s claim that billions cannot be earned ethically, citing a founder with 93% monthly revenue growth whose net worth increases proportionally. The core argument is that extreme wealth in startups stems from exponential growth, not cheating, making the path reproducible albeit difficult.

Jun 15, 2026paulgraham.com

GameBoy Workboy

The unreleased Game Boy accessory Workboy, intended as a micro workstation with keyboard functionality, was recovered as a ROM from the September 2020 Nintendo leak. The ROM contains unused text and displays a version number discrepancy: the title screen shows 8.87, while internal strings reference version 5.74. A prototype keyboard surfaced in December 2020 and was

Jun 14, 2026tcrf.net

Free SQL→ER diagram tool, runs in the browser, nothing uploaded

SQL to ER Diagram is a free, open-source tool that parses SQL CREATE TABLE and ALTER TABLE statements into interactive entity-relationship diagrams entirely within the browser, with no server uploads or sign-up required. It supports PostgreSQL, MySQL, SQLite, and SQL Server syntax, including primary and foreign keys, unique, and not-null constraints. Users can drag tables to rearrange the layout, zoom, add notes, and export the resulting diagram as PNG or SVG.

Jun 14, 2026sqltoerdiagram.com

ReactOS (FOSS "Windows") achieves 3D-accelerated Half-Life on real hardware

ReactOS, the open-source operating system designed for binary compatibility with Windows applications and drivers, has achieved a milestone by successfully running the Windows version of Half-Life on real hardware. User "Zombiedeth" demonstrated gameplay on a Dell OptiPlex with a Core i5-2400 (Sandy Bridge) and NVIDIA GeForce 8400GS. While Half-Life already runs well on Linux via Wine, this marks the first reported instance of

Jun 14, 2026phoronix.com

Jun 14, 2026

AI OSS tool repo goes archived over night after raising $7.3M Seed

TensorZero, an open-source LLMOps platform unifying an LLM gateway with observability, evaluation, optimization, and experimentation, has had its GitHub repository archived overnight after raising a $7.3 million seed round. The repo’s

Jun 14, 2026

AI coding at home without going broke

The article outlines three strategies for cost-effective AI coding at home: self-hosting open-source models (steep upfront cost, weaker performance), renting open models via API providers like OpenRouter (flexible, avoids hardware lock-in), and leveraging metered frontier subscriptions from OpenAI/Anthropic (~$400/month yielding ~$2,800 in API value, but capped). The

Jun 14, 2026

Police officer investigated for using AI to 'create evidence' in multiple cases

A police officer is under investigation for allegedly using generative AI to fabricate evidence across multiple cases, raising serious concerns about forensic integrity and chain-of-custody protocols. The misconduct reportedly involved creating false documentation or digital artifacts that could compromise criminal proceedings and erode trust in AI-assisted investigative tools. This case highlights emerging risks in law enforcement as generative models become accessible, emphasizing the need for strict audit trails and validation of AI-generated outputs in legal contexts.

Jun 14, 2026

GLM 5.2 Is Out

Zhipu AI released GLM-5.2 as its most capable fully open-source model, featuring a usable 1M context window and a demonstrated lead in independent long-horizon task completion for complex agent applications. The model serves as the primary engine for Zhipu's strongest domestic coding model and became available to all GLM Coding Plan users (

Jun 14, 2026

Treating pancreatic tumours may have revealed cancer's master switch

A novel treatment approach for pancreatic tumours has potentially identified a single regulatory mechanism that controls cancer growth and spread. This "master switch" could explain how tumours evade treatment and metastasize, offering a unified target for therapy. The finding suggests that disrupting this pathway might stop cancer progression across multiple tumor types

Jun 14, 2026

US bans differential privacy in Census data

The U.S. Department of Commerce has ordered a ban on "noise infusion"—the addition of random noise to statistical outputs—from all products published by the Census Bureau and Bureau of Economic Analysis. This technique is a core component of differential privacy, widely considered the gold standard for disclosure avoidance, as it provably limits how much a single individual

Jun 14, 2026

Every Frame Perfect

Wayland’s stated goal of “every frame is perfect” is extended to UI design: any screenshot of an application must be visually coherent, with no white flashes, partially loaded content, layout shifts, or desynchronized animations. Such imperfections erode user trust, as UI quality is the primary heuristic for judging overall code quality. The argument applies precise technical scrutiny to visual states, treating transient frames as first-class deliverables rather than afterthoughts.

Jun 14, 2026

Israeli firm BlackCore suspected of meddling in New York and Scotland votes

An Israeli cybersecurity firm, BlackCore, is suspected of orchestrating influence operations targeting elections in New York and Scotland. The alleged meddling likely involved coordinated disinformation campaigns, bot networks, or other technical manipulation of online platforms. Reuters reports this as a developing story, highlighting growing concerns over private-sector involvement in foreign election interference.

Cybersecurity

Latest Cybersecurity

All Cybersecurity

Arch Linux AUR Hit by Another Wave of Now More Sophisticated Malware Attack

A second wave of malware targeting Arch Linux AUR packages has been discovered, using code obfuscation to conceal its intent in contrast to the earlier incident affecting over 1,500 packages. Affected packages include Node.js packages, a Plasma 6 applet, Firefox packages, the Aura browser, LibreWolf extensions, and a NeoVim plugin. Developer a821 identified the initial batch, and Nicolas Boichat subsequently found additional obfuscated malware using

Jun 15, 2026phoronix.com

Apt Encounters of the Third Kind

During a routine security assessment of a client’s cluster of reverse gateways—each running a custom Linux stack with a monolithic kernel and a GoLang application booting from USB—an NFS server was found to contain two malicious binaries. Forensic analysis revealed that the kernel had been patched, while the GoLang application itself remained unmodified, leading to the discovery of a sophisticated APT-style compromise targeting the NFS-mounted configuration and TLS data. The investigation uncovered that the attackers exploited the hardcoded NFS mount in the init-replacement Go binary to deliver persistent malware.

Jun 14, 2026igor-blue.github.io

France's Own Hack Is the Best Argument Against Its War on Encryption

France's government-built secure messenger Tchap, used by civil servants, was breached on June 7 via credential theft rather than any cryptographic failure—neither the encryption nor servers were broken. The attacker hijacked a legitimate user account, demonstrating that weak access control (not encryption) was the vulnerability. This incident directly undermines France's push for encryption-backdoor proposals like Chat Control, as it shows that compromising user credentials is a simpler attack vector than bypassing strong encryption, undercutting claims that state access to message scanning is necessary or safe.

Jun 14, 2026reclaimthenet.org

Arch Linux Now Believes Malware Incident Under Control: More Than 1,500 Packages

Arch Linux developers have contained a malware incident affecting the Arch User Repository (AUR), with over 1,500 user-contributed packages compromised. The number of affected packages escalated from an initial 400 to roughly 900, then to 1,579, though the final list likely does not capture all impacted packages. All known malicious commits have been deleted.

Jun 13, 2026phoronix.com

Open Source AI Must Win

Open-source AI is framed as a civilizational infrastructure requiring operational freedom—the ability to study, deploy, audit, and modify models without permission—rather than mere software freedom. The argument warns that closed frontier labs and platform companies risk reducing AI access to a subscription economy for cognition, dependent on opaque APIs, shifting terms, and centralized control. The practical posture advocated is American capacity combined with global open standards, ensuring models remain locally deployable, reproducible, and community-governed regardless

Jun 13, 2026opensourceaimustwin.com

AMD Stiffs Researcher $10k Bug Bounty

AMD's auto-updater downloaded software over unencrypted HTTP, allowing remote code execution via man-in-the-middle attacks. The company acknowledged the critical vulnerability but denied the researcher's $10,000 bug bounty payment. The fix required 124 days and still relies on weak CRC32 validation rather than cryptographic signature verification.

Jun 13, 2026gadgetreview.com

Jun 13, 2026

Twenty One Zero-Days in FFmpeg

A depthfirst autonomous security agent discovered 21 zero-day vulnerabilities in FFmpeg, including issues that had remained latent for 15–20 years despite prior analysis by Google and Anthropic. The agent generated concrete, reproducible proof-of-concept inputs to confirm each finding, achieving a per-finding cost of roughly $1k compared to $10k for traditional approaches. The team explored exploit

Jun 13, 2026

Eric S. Raymond: How to Become a Hacker

Eric S. Raymond's essay defines a hacker as a person who delights in solving problems and pushing limits, not as a security cracker. It outlines five essential attitudes: curiosity, rejection of boredom, willingness to share, critical thinking, and a belief that the world can be improved through technology. Raymond prescribes learning C, Lisp, Python, and Unix basics, while emphasizing hands-on practice with open-source projects. The document remains a foundational cultural

Jun 13, 2026

Arch Linux's AUR Sees More Than 400 Packages Compromised with Malware

Over 400 user-supplied packages in the Arch Linux User Repository (AUR) were compromised in a large-scale malware campaign this week. Arch Linux maintainers have been resetting or deleting all malicious content and banning the affected accounts. The attack is confined to the AUR and does not impact official Arch Linux packages. Further details and discussion are available on the Arch Linux mailing list and CachyOS forums.

Jun 13, 2026

Law Enforcement's "Warrior" Problem (2015)

Law enforcement training and culture have extensively adopted a "warrior mindset,"

Jun 13, 2026

Malware developers added nuclear and biological weapons text to to their spyware

Malware developers have embedded references to nuclear and biological weapons within spyware code, likely as a tactic to hinder automated analysis or evade signature-based detection. The inclusion of sensitive terminology may also serve as a false flag to mislead incident responders or escalate perceived threat levels.

Jun 13, 2026

Kimi K2.7-Code: open-source coding model with better token efficiency

Moonshot AI has released Kimi-K2.7-Code, an open-source coding model optimized for improved token efficiency. The model is hosted on Hugging Face and is designed to reduce computational overhead while maintaining performance on code generation tasks. This release targets developers and AI researchers seeking more cost-effective code completion and generation solutions.

Jun 12, 2026

Why I'm Forced to Say Farewell: Google Management Has Lost Its Moral Compass

René Mayrhofer, former Director of Android Platform Security at Google, announced his departure due to a perceived loss of moral integrity in Google’s management. He joined in 2017, drawn by Android’s open-source ethos and its role as the most widely used Linux-based OS, where he led security mitigations and mobile digital ID architectures. Mayrhofer contrasts the transparent, academically-al

Jun 12, 2026

Who Runs the Ransomware Group 'The Gentlemen?'

A cybercrime group called The Gentlemen has become the second most active ransomware operation by victim count, using a 90/10 affiliate revenue split (vs. industry standard 80/20) to attract experienced operators from competing programs. Check Point researchers report the group targets internet-facing devices such as VPNs and firewalls, then encrypts entire networks within hours. The administrator, who uses the nicknames Zeta88 and Hastalamuerte on Russian-language forums, was identified after the group’s backend infrastructure was breached.

AI & LLM

Top AI News

All AI News

Rio de Janeiro's "homegrown" LLM appears to be a merge of an existing model

Rio-3.5-Open-397B is a merged model composed of roughly 60% Nex-N2_pro and 40% Qwen, as disclosed in a GitHub issue. The model’s claimed “homegrown” nature is contradicted by this explicit composition, indicating it is a

Jun 15, 2026github.com

Cloud-based LLM gold rush is ending

Apple's WWDC 2026 announcement signals a strategic shift from cloud-based LLMs to local AI processing on MacOS, effectively marking the end of the LLM gold rush. The company argues that most user workflows and automations can run natively on-device

Jun 15, 2026automato.substack.com

Rio de Janeiro's city government model Rio3.5 beats Qwen3.7 in recent benchmarks

Alibaba's Qwen3.7 is losing frontier relevance due to its proprietary stance, as open models like Minimax M3 and Rio 3.5 397B—a 397-billion-parameter model developed by the

Jun 15, 2026twitter.com

EU Commission looking at practical consequences of Anthropic decision

The European Commission is evaluating the real-world implications of a recent regulatory or legal decision concerning AI company Anthropic. This review focuses on how the ruling might affect compliance obligations, market operations, and enforcement mechanisms under the EU’s evolving AI regulatory framework. The outcome could set precedents for how foundational model developers are treated under the AI Act and related digital policy. Technical stakeholders should monitor this as it may clarify liability and transparency requirements for large language models operating in Europe.

Jun 15, 2026reuters.com

The whirlwind 24 hours that led to export controls on Anthropic

The U.S. government imposed export controls on Anthropic within a 24-hour period, targeting the AI

Jun 14, 2026politico.com

LLMs aren't conscious (and thinking they are is culturally dangerous)

Pope Leo XIV's encyclical *Magnifica Humanitas* asserts that AI systems, including LLMs, are not conscious because they lack subjective experience, embodiment, moral conscience, and genuine understanding—a position author Erik Hoel endorses. Hoel argues that while LLMs can simulate expressions of love or empathy, they possess no intrinsic, first-person perspective; outputting

Jun 14, 2026theintrinsicperspective.com

Jun 14, 2026

Don't trust large context windows

LLM context windows have a "smart zone" up to roughly 100K tokens and a "dumb zone" beyond it, where attention drops and the model forgets earlier inputs. Studies such as RULER and Chroma’s report on context rot confirm that effective context is a fraction of advertised windows (200K–2M), making those numbers primarily

Jun 14, 2026

Making Claude a Chemist

Anthropic is collaborating with synthetic, computational, and analytical chemists to enhance Claude’s proficiency in chemistry, beginning with NMR spectrum interpretation—a core analytical tool. The work targets the model’s ability to translate between chemical representations (e.g., drawn structures, spectral readouts, database query strings) that encode the same molecule but require distinct forms of fluency. Accurately parsing these signals is critical, as small structural changes can drastically alter a compound’s

Jun 14, 2026

RTX 5080 and RTX 3090 Setup: 80 Tok/s on Qwen 3.6 27B Q8

Achieving over 80 tokens per second on Qwen 3.6 27B at Q8 quantization requires splitting a PCIe 4.0 x16 slot into two x8 lanes, with the Asus Prime X570-Pro specifically supporting this via its BIOS (disable CSM,

Jun 13, 2026

Shepherd's Dog: A Game by the Most Dangerous AI Model

Anthropic's recently released AI model, previously deemed too dangerous for public release, generated a complete, playable game called "Shepherd's Dog" in a single 45-minute reasoning session costing over €20 in tokens. The model produced a standalone 2,319-line index.html file with zero external dependencies, implementing a game concept the author had held for years. This marks the first time an AI has created a fully functional game from a single prompt, demonstrating

Jun 13, 2026

I Think They [Anthropic] Are Lying to You [video]

A YouTube video titled "I Think They [Anthropic] Are Lying to You" alleges that Anthropic, the

Jun 13, 2026

We've suspended access to Claude Mythos 5 and Claude Fable 5

Anthropic suspended access to Claude Mythos 5 and Claude Fable 5 as of June 13, 2026, impacting all service endpoints including claude.ai, the Claude API (api.anthropic.com), Claude Code, and Claude Cowork. The incident is currently under monitoring, with further details available at Anthropic's official blog post. No root cause or estimated resolution time has been provided.

Jun 13, 2026

Trump admin blocks foreign access to Anthropic's most powerful AI models

The Trump administration has imposed restrictions preventing foreign entities from accessing Anthropic's most advanced AI models, likely Claude 3.5 Sonnet or Opus, under national security and export control authorities. This move extends existing chip-level export

Jun 13, 2026

US Government directive to suspend access to Fable 5 and Mythos 5

The US government issued an export control directive on June 12, 2026, suspending all access to Anthropic's Fable 5 and Mythos 5 models for any foreign national, including