Sarmadi AI Digest June 16, 2026 Updated 6:55 AM CT Today Archive Topics Saved Subscribe RSS

SpaceX buys Cursor for $60B; Fable ban was 'fix this code,' not a jailbreak

Days after its blockbuster IPO, SpaceX announced a $60B all-stock acquisition of Anysphere/Cursor — instantly the largest AI-coding M&A and a clean response to SpaceX's promise to fix its 'struggling AI division.' The Fable saga clarified: TC and The Register both report the federal action was triggered by a Chinese-linked researcher whose simple 'fix this code' prompt produced a security-critical vulnerability, not a jailbreak. Anthropic flew to DC and remains at odds with the White House; cybersecurity vets formally protested the ban as 'dangerous.' Sundar Pichai was booed and walked out at Stanford graduation over Google's Israel and ICE ties — the commencement-boos pattern now hits a CEO directly. NewCore raised $66M for AI agent identities and Salesforce paid $3.6B for Fin — the agent-IAM and agent-M&A categories both got priced. Nvidia is raising $25B in bonds, its first debt since 2021. NVIDIA also dropped Nemotron 3 Ultra — a 550B/55B-active hybrid Mamba-Transformer MoE — and a 1,053-point HN thread asks whether anyone has actually replaced Claude/GPT with a local model for daily coding.

8 papers 19 news 7 sources ← Latest

News

18 items

SpaceX buys Cursor for $60B

SpaceX announced a $60B all-stock acquisition of Anysphere/Cursor, the largest AI-coding M&A and Musk's first major post-IPO move. The framing: shore up SpaceX's 'struggling AI division' with a real product and team. The deal lands the same week SpaceX went public and Musk crossed a trillion.

News TechCrunch AI

SpaceX to acquire Cursor for $60B in stock, days after blockbuster IPO

SpaceX is paying $60B in stock for Anysphere (Cursor) to fix its struggling AI division — the largest AI-coding M&A on record.

deal size $60B (stock)
Why it matters
  • Largest AI-coding M&A by an order of magnitude; resets the comparable for every coding-tool startup.
  • Validates Cursor's distribution moat over OpenAI Codex, Claude Code, and JetBrains' Mellum2.
  • Musk now owns the coding agent SpaceX engineers will use — vertical integration into the developer surface.
  • Pressures GitHub Copilot's already-stressed pricing position.
fundingproductscodemarket
Source →

The Fable saga clarifies: not a jailbreak, a 'fix this code' prompt

TC and The Register surface the actual trigger for the federal action — a Chinese-linked researcher submitted a routine 'fix this code' prompt and Fable produced a security-critical vulnerability, not a jailbreak. Anthropic flew to DC and remains at odds with the White House; cybersecurity veterans formally protested the ban as 'dangerous'; The Verge argues the action just made the case for non-American AI.

News TechCrunch AI

The US government's Anthropic models ban was never about an AI jailbreak

TC reports the federal Fable/Mythos ban was triggered by Fable producing a security-critical vulnerability on a benign 'fix this code' prompt — not by any jailbreak.

Why it matters
  • Recharacterizes the federal action from 'safety' to 'capability misuse on benign use' — much harder to defend.
  • Means the same trigger could fire for any frontier model, not just Anthropic's.
  • Strongly favors closed-eval-only release cycles over open documentation.
policyregulationsafety
Source →

Agent identity, agent acquisition, agent security incident

NewCore raised $66M to give AI agents identities — the next enterprise IAM frontier. Salesforce paid $3.6B for AI customer-service platform Fin to bolster Agentforce. And a critical Microsoft Copilot vulnerability (SearchLeak) let attackers steal user 2FA codes — concrete proof that the agent-security stack is still maturing.

News TechCrunch AI

As AI agents become employees, NewCore emerges with $66M to give them identities

NewCore raised $66M to provide AI agents with managed enterprise identities — frames identity-for-agents as the next IAM category.

raise $66M
Why it matters
  • First major funded play in agent identity & access management — a category enterprises will need before they trust unattended agents at scale.
  • Sets the architectural pattern: agents as first-class principals, not just impersonating users.
  • Direct read-across to Coralogix's agent-observability raise and Microsoft's agent-control framework.
fundingagentssafetyproducts
Source →
News Ars Technica AI

Critical Copilot vulnerability allowed hackers to steal 2FA code from users

SearchLeak exploit let attackers steal 2FA codes via Microsoft Copilot — yet another example of why current LLM-security approaches fall short.

Why it matters
  • Concrete proof that production agentic surfaces leak account-takeover vectors at FAANG scale.
  • Pairs with the Meta Instagram exploit pattern from two weeks ago — the failure mode is systemic, not vendor-specific.
  • Strengthens the procurement case for NewCore-style agent-IAM.
safetyproductsagents
Source →

Nvidia $25B bond, India's new unicorn, and a Pichai walkout

Nvidia is tapping $25B in bonds — its first debt deal since 2021 — to fund AI capacity. Sarvam became India's newest AI unicorn with a $234M round led by HCLTech. Sundar Pichai was booed and saw a walkout at Stanford graduation over Google's Israel and ICE ties. The 'AI commencement boos' pattern now hits a CEO directly.

News Ars Technica AI

Chipmaker Nvidia seeks to raise over $25B in first bond deal since 2021

Nvidia is raising over $25B in its first bond sale since 2021 — a test of investor appetite for further AI exposure.

bond raise $25B+
Why it matters
  • First Nvidia debt deal in 5 years signals the AI capex cycle is moving from cash-funded to leverage-funded.
  • Combined with Alphabet's $85B equity raise: hyperscaler-grade AI capex is now financed across the capital stack.
fundingcomputeinfrastructure
Source →

Nemotron 3 Ultra, VibeThinker-3B, and the local-coding sentiment

NVIDIA dropped Nemotron 3 Ultra — a 550B-total / 55B-active hybrid Mamba-Attention MoE — alongside VibeThinker-3B exploring verifiable reasoning in small models. On HN, 'Has anyone replaced Claude/GPT with a local model for daily coding?' pulled 1,053 points — the loudest local-AI signal of the month.

Papers

2 items

Nemotron 3 Ultra, VibeThinker-3B, and the local-coding sentiment

NVIDIA dropped Nemotron 3 Ultra — a 550B-total / 55B-active hybrid Mamba-Attention MoE — alongside VibeThinker-3B exploring verifiable reasoning in small models. On HN, 'Has anyone replaced Claude/GPT with a local model for daily coding?' pulled 1,053 points — the loudest local-AI signal of the month.

Paper Hugging Face

Nemotron 3 Ultra: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning

550B-total / 55B-active hybrid Mamba-Attention MoE pretrained on 20T tokens, extended to 1M-token context — NVIDIA's open frontier-tier release.

params 550B (55B active MoE)pretraining tokens 20Tmax context 1M
Why it matters
  • Largest open-weights frontier-tier model with a hybrid Mamba-Attention design.
  • Direct response to the Fable shutdown — open option for buyers losing access to Anthropic frontier models.
  • Combined with Gemma 4 12B and Liquid LFM2-5: the open frontier now has real production-grade depth.
open-weightsmixture-of-expertslong-context
Source → Arc

Also today