X/Twitter AI Pulse — 2026-05-13

Top AI Discussions This Week

Anthropic's Near-Trillion-Dollar Fundraise Stuns the Tech World

• Who's talking: AI investors, researchers, and tech media across X/Twitter
• What happened: The New York Times reported that Anthropic is in talks to raise funding at a $950 billion valuation — more than doubling its previous valuation of $380 billion. The company recently released a powerful model called "Mythos" and is separately embroiled in a dispute with the Pentagon.
• Key takes: The community is reacting with a mix of awe and skepticism. Forbes noted that Anthropic now leads OpenAI in business growth with a reported $44B ARR and 70% margins, despite OpenAI's $852B valuation. Debates are swirling about whether these valuations reflect sustainable fundamentals or speculative excess — and when either company will actually IPO.
• Why it matters: A $950B AI startup valuation would be historic, signaling that capital markets still believe frontier AI labs are among the highest-value assets on earth, even amid ongoing cost and profitability questions.

2 sources

t.co

t.co

t.co

t.co

AI Used for First-Ever Autonomous Zero-Day Exploit Discovery

• Who's talking: Security researchers, AI safety advocates, and developers on X/Twitter
• What happened: Reuters reported on May 11 that Google confirmed hackers from a prominent cybercrime group used AI to discover a previously unknown software vulnerability and develop an exploit autonomously — the first known instance of this capability being used in the wild.
• Key takes: The AI security community is alarmed. Many researchers note this validates long-standing warnings about offensive AI capabilities accelerating beyond defensive ones. The Air Street Press "State of AI: May 2026" report had already flagged that frontier cyber-offense capability is "doubling every four months."
• Why it matters: This marks a qualitative shift in the threat landscape. AI-assisted zero-day discovery had been a theoretical concern; it is now a confirmed real-world attack vector, likely to intensify regulatory and industry pressure around dual-use AI.

1 sources

reuters.com

reuters.com

EU Presses OpenAI and Anthropic for Direct Model Access

• Who's talking: EU regulators, AI policy watchers, enterprise AI users
• What happened: The European Commission formally pressed OpenAI and Anthropic for direct AI model access, seeking safety reviews of GPT-5.5-Cyber and Mythos, and asserting EU oversight authority under emerging AI regulations.
• Key takes: Observers on X note this could set a precedent for governments demanding "keys" to frontier AI systems. Some argue this is a reasonable safety mechanism; others warn it creates backdoors and geopolitical leverage. The timing — as both companies are reportedly raising enormous new rounds — adds political complexity.
• Why it matters: This is one of the first concrete enforcement-adjacent moves by a major government under new AI oversight frameworks, and it could reshape how frontier models are deployed in Europe.

Hot Debates & Controversies

Zvi Mowshowitz and the Community Debate Claude Opus 4.7's Benchmarks

• Side A: @TheZvi and others argue Claude Opus 4.7 represents a meaningful capability leap — notably, its training cutoff jumped from May 2025 (Opus 4.6) to end of January 2026, and it has taken the #1 spot on Artificial Analysis benchmarks. The improved knowledge currency is seen as a "big practical deal."
• Side B: Skeptics counter that benchmark leadership is increasingly fleeting — models leapfrog each other every few months — and question whether marginal benchmark gains translate to real-world task performance.
• Current status: The debate is ongoing. Community consensus seems to be that Opus 4.7 is a solid upgrade, but the broader war for model supremacy remains wide open.

Perceptron Mk1: Cheap Video AI vs. Big Lab Pricing

• Side A: Perceptron Mk1 has generated buzz by claiming its video analysis model is 80–90% cheaper than Anthropic, OpenAI, and Google alternatives, while delivering competitive performance. Early adopters report using it for auto-clipping live sports highlights using temporal scene understanding.
• Side B: Skeptics question whether the cost savings hold at scale and whether a smaller lab can sustain model quality and reliability against well-resourced incumbents. Others raise concerns about long-term enterprise support.
• Current status: Still early days, but the announcement is forcing a community conversation about whether frontier pricing by big labs is justified — or whether cost compression is coming faster than expected.

Notable AI Announcements

• White Circle (French startup): Raised $11M in seed funding backed by operators from OpenAI, Anthropic, DeepMind, Hugging Face, Mistral, Datadog, and Sentry to build enterprise AI monitoring and security tools — community reaction is enthusiastic, noting the all-star backer list signals this is a serious problem worth solving.

• Anthropic (Mythos + $950B valuation talks): In addition to the funding talks, Anthropic's recently released "Mythos" model is drawing attention as the apparent driver of its accelerating ARR and margin story — community reaction ranges from impressed to questioning whether the valuation math holds long-term.

• Sequoia Capital (AI Ascent 2026): Sequoia posted highlights from its AI Ascent 2026 event featuring talks with Andrej Karpathy, Demis Hassabis, Jim Fan, and others — community reaction has been positive, with many clipping and sharing Karpathy's segment in particular.

Thought Leader Spotlight

@TheZvi on Claude Opus 4.7 Capabilities

• Key quote/insight: Zvi highlighted that Opus 4.7's training cutoff moved from May 2025 to end of January 2026 — calling it "a big practical deal" — and noted the model now holds the #1 position on Artificial Analysis benchmarks (in a tie).
• Context: Posted as a detailed two-part breakdown of Opus 4.7 capabilities and community reactions to the model's release.
• Community reaction: The knowledge cutoff point resonated strongly — many users noted that an 8-month improvement in training data currency matters enormously for real-world use cases involving current events, regulations, and market data.

@DataChaz on Karpathy's AI Agent Playbook

• Key quote/insight: Referencing Andrej Karpathy's widely circulated advice, the post states: "Karpathy was right. He warned that 90% of AI advice dies in 6 months. Most tools will not even survive 90 days." The thread frames 2026 as the year to focus on durable AI agent fundamentals, not hype-driven tooling.
• Context: Prompted by the rapid churn of AI tools and the Sequoia AI Ascent 2026 event featuring Karpathy.
• Community reaction: Widely reshared, with many developers expressing relief at having a framework for cutting through tool-of-the-week noise. Others debated whether "AI whispering" is actually a durable skill or itself a transitional phase.

What to Watch Next Week

• Anthropic funding round close: Watch for confirmation (or complications) on the reported $950B valuation raise, which would be the largest private AI financing event in history.
• EU AI model access negotiations: The European Commission's demand for direct access to GPT-5.5-Cyber and Mythos is a developing regulatory story — any formal response from OpenAI or Anthropic could set major precedents.
• AI-enabled cyberattacks in the wild: Following Google's confirmation that AI was used to discover a zero-day exploit, security researchers and policymakers are expected to respond — watch for new guidance, disclosures, or emergency regulatory action.