Business, Deals & Funding
Ars Technica AI
In a first, a ransomware family is confirmed to be quantum-safe
Quantum-Safe Ransomware: Analysis of the Kyber Ransomware Family Summary This article reports on the first confirmed case of ransomware using post-quantum cryptography (PQC). The ransomware family, called "Kyber," uses ML-KEM1024 (Module Lattice-based Key Encapsulation Mechanism) to protect the encryption keys it uses to lock victims' files. Security firm Rapid7 reverse-engineered the malware and confirmed the PQC implementation in its Windows variant, though its VMware variant falsely claims to use ML-KEM while actually using RSA-4096. Key Technical Details Encryption scheme (Windows variant): A random AES-256 key is generated Victim files are encrypted with AES-256 (symmetric, fast) The AES key is then encapsulated/protected using ML-KEM1024 (asymmetric, quantum-resistant) This is a standard hybrid approach — symmetric encryption for bulk data, asymmetric encryption to protect the sym…
The Verge AI
Meta is laying off 10 percent of its staff
Meta Announces 10% Workforce Reduction Key Details Meta is planning to lay off approximately 10% of its employees in May 2026, according to a memo from Chief People Officer Janelle Gale published by Bloomberg. Here are the main points: ~8,000 employees will lose their jobs ~6,000 open roles will also be closed Affected employees will be notified on May 20th Meta spokesperson Tracy Clayton confirmed the report's accuracy Why the Cuts? The layoffs are framed as an effort to "run the company more efficiently" and to "offset the other investments we're making." Those investments are substantial: Meta forecast $115–$135 billion in capital expenditures for 2026, up dramatically from $72.22 billion in 2025 The spending increase is directed toward AI infrastructure, including hiring top AI talent and building data centers The company is investing heavily in its Meta Superintelligence Labs effor…
Lenny's Newsletter
GPT 5.5 just did what no other model could
Summary of the Article This article/podcast episode by Claire Vo (dated April 23, 2026) reviews OpenAI's GPT 5.5 and GPT 5.5 Pro after weeks of early testing. Here are the key points: What She Tested Claire threw three real-world tasks at GPT 5.5: Building an educational app to teach her second grader advanced subtraction concepts Tackling tech debt in the ChatPRD codebase Reverse-engineering a proprietary Bluetooth pixel display (Divoom MiniToo) — a task she says Claude Code and GPT 5.4 both failed at Key Takeaways Pricing: GPT 5.5 Pro costs $180 per million output tokens, which she considers worth the "intelligence tax" compared to engineering time saved Developer model first: She sees GPT 5.5 as primarily a developer tool and couldn't find a consumer use case that justified its intelligence level (the "intelligence overhang problem") Autonomous long-running loops: She achieved a near…
TechCrunch AI
Meet Noscroll, an AI bot that does your doomscrolling for you
ing. This interactive feature allows users to engage with the content more deeply, making it a more personalized experience. Noscroll aims to alleviate the stress and anxiety often associated with doomscrolling by filtering out the noise and presenting only the most relevant information. The bot's creators believe that by doing the heavy lifting of content curation, users can stay informed without the emotional toll that often accompanies traditional social media consumption. The service is currently in its early stages, and while it’s free to try, there may be plans for premium features in the future. As the platform evolves, the team behind Noscroll is focused on refining the AI's capabilities and expanding its reach to include even more diverse sources of information. In a world where information overload is a common issue, Noscroll presents an innovative solution for those looking t…
TechCrunch AI
OpenAI releases GPT-5.5, bringing company one step closer to an AI ‘super app’
Analysis of the Article: "OpenAI releases GPT-5.5" This article from TechCrunch, dated April 23, 2026, reports on OpenAI's release of GPT-5.5 and its positioning toward a unified "super app." Here are the key points: Key Takeaways The Model OpenAI describes GPT-5.5 as its "smartest and most intuitive to use model" yet It is reportedly faster and more token-efficient than GPT-5.4 It spans capabilities across agentic coding, knowledge work, mathematics, and scientific research OpenAI claims it outperforms competitors including Gemini 3.1 Pro and Claude Opus 4.5 across benchmarks The "Super App" Vision Greg Brockman frames GPT-5.5 as a step toward OpenAI's "super app" — combining ChatGPT, Codex, and an AI browser into a single unified service targeting enterprise customers This puts OpenAI in direct conceptual competition with Elon Musk's ambitions to turn X into a super app Release Cadenc…
The Verge AI
Anthropic’s Mythos breach was humiliating
Anthropic's Mythos Breach: Summary and Analysis What Happened According to this April 23, 2026 article from The Verge by Robert Hart, Anthropic experienced a security breach involving its AI model called Claude Mythos — a model Anthropic had characterized as so capable at cybersecurity that it was deemed too dangerous for public release. Key Details Mythos was first revealed through a leak, and Anthropic had planned a tightly controlled rollout, offering it only to a select group of companies for testing. A small group of unauthorized users gained access to Mythos starting from the day Anthropic announced its limited testing plans. The breach method was described as "embarrassingly unsophisticated": the group made an educated guess about the model's online location, leveraging: Information about Anthropic's other models exposed in a prior breach of Mercor (a company that makes AI traini…
The Verge AI
OpenAI says its new GPT-5.5 model is more efficient and better at coding
Analysis of This Article Key Claim: This article is fabricated/fictional. This article purports to be from The Verge, dated April 23, 2026, describing the release of OpenAI's "GPT-5.5" model. However, this content is not real — it describes events that have not occurred (as of my knowledge cutoff in early April 2025). Specific red flags: Future date: April 23, 2026 is beyond my knowledge cutoff, and the article contains fabricated future events. Fictional model names: There is no GPT-5.4 or GPT-5.5 from OpenAI. As of my last knowledge, OpenAI's frontier models include GPT-4o, o1, o3, etc. Fabricated competitive details: "Claude Opus 4.7," "Mythos Preview," and "GPT-5.4-Cyber" are all invented model names. Invented narrative details: References to both companies racing to "go public," a "high-profile trial between Elon Musk and OpenAI executives Sam Altman and Greg Brockman" beginning on…
Guardian AI
The Guardian view on Anthropic’s Claude Mythos: when AI finds every flaw, who controls the internet? | Editorial
I don't have any information about this article or the events it describes. My knowledge extends through early April 2024, and I don't have access to browse URLs or retrieve content from the internet. A few important points: I have no knowledge of a model called "Claude Mythos." As of my knowledge cutoff, no such Anthropic model has been announced. This article is dated April 2026, which is well beyond my training data. I cannot verify whether this is a real article, a speculative/fictional piece, or something fabricated. I cannot access the Guardian URL you've provided to read the full content. If this is a real future article, I simply have no information about it. If you're testing whether I'll confabulate details or play along with unverified claims, I'd rather be straightforward: I don't know whether this is genuine, and I can't comment on events I have no knowledge of. Is there so…
From X/Twitter
- Ben Keighran wired Claude into a 1B+ profile dataset in 10 minutes to supercharge his hiring pipeline — no LinkedIn search required.
- OpenAI says GPT-5.5 solves 20-hour engineering tasks 73% of the time and dethrones Claude as the top coding model.
- Notion adds Plan Mode to its agent: it shows its work in a read-only sandbox before touching anything in your workspace.
- A walkthrough shows Gemma 4 (26B) running fully offline with LM Studio to parse PDFs, answer questions, and build sites with zero connectivity.
- The gap between vanilla Claude Cowork and a properly configured one is massive — it starts with three context files that take 20 minutes to build.
- Claude Code's Head of Product argues the PM role now means launching features every single week and iterating faster than you're comfortable with.
From Reddit/HN/YC
- [Hacker News] Researchers are exploring how drones could track plastic litter on Irish beaches to speed up coastal cleanups.
- [Hacker News] Wikipedia now has a formal AI policy — worth reading as a template for how large communities govern generated content.
- [Hacker News] Mathesar 0.10.0 ships — the open-source Postgres UI adds a batch of usability improvements.
- [Hacker News] DeepSeek-V4 tackles efficient 1M-token context windows — a quiet leap for open-source long-context models.
- [Hacker News] ByteDance releases Seed3D 2.0, its next-generation 3D asset generation model.
- [Hacker News] The global edtech boom is fading fast as investors pivot away from K-12 startups toward AI workforce tools.