The AI Brief

Space Exploration Technologies Corp. agreed to acquire Anysphere, the parent company of AI coding tool Cursor, in an all-stock deal valuing the startup at $60 billion. SpaceX expects the transaction to close during the third quarter of 2026, pending regulatory approvals. The deal was announced at Cursor's inaugural Compile conference in San Francisco on June 16.

SpaceX announced an option in April to buy Cursor for $60 billion or pay a $10 billion break-up fee; that option converted into a signed merger agreement on June 16. At the time SpaceX announced the option, Cursor was in the middle of raising a $2 billion funding round that would have valued the company above $50 billion, with Andreessen Horowitz, Nvidia, and Thrive Capital set to participate; the $60 billion deal price exceeded that figure. Microsoft had examined a potential acquisition of Cursor before the SpaceX arrangement emerged but ultimately decided against submitting a formal bid; Cursor had also rebuffed two separate approaches from OpenAI.

According to Reuters, Cursor generates approximately $2.6 billion in annualized B2B revenue; per Dealroom, total ARR has reached roughly $4 billion. The deal comes four days after SpaceX's Nasdaq IPO and is designed to strengthen the AI division built around xAI, which merged with SpaceX in February 2026. Cursor's market share had declined from 41% in June 2025 to about 26% in May, according to spending data from Ramp; Anthropic now controls half of that category. Separately, at the same Compile event, Cursor announced Origin on June 17, a git hosting and code collaboration platform with a single provocative premise: what if the primary users of your version control system are no longer human? Origin's underlying technology comes from Cursor's acquisition of Graphite, optimised specifically for the high-frequency, parallel code operations characteristic of AI agents, with a built-in AI-powered automatic merge conflict resolution engine. Origin is positioned as a direct GitHub competitor, targeting a fall 2026 launch.

Morgan Stanley forecasts AI-linked global debt issuance will reach nearly $570 billion in 2026, more than doubling, according to a report published on June 10. The brokerage cited rising bond supply and increased credit market activity as hyperscalers seek alternative funding sources to meet massive capital expenditure demands driven by AI. Morgan Stanley estimated that AI-related global debt issuance stood at nearly $236 billion as of May 31, 2026, a fourfold increase compared to the same period in the previous year.

Hyperscaler capital expenditures in 2026 are on pace to consume close to 100% of operating cash flows, compared with a 10-year average of 40%, according to UBS. Tech companies have reshaped the U.S. investment-grade bond market: Amazon, Alphabet, Meta, Microsoft, and Oracle collectively issued $121 billion in U.S. corporate bonds in 2025, compared with an average of $28 billion per year between 2020 and 2024, according to Bank of America. By October 2025, AI-linked debt had reached $1.2 trillion, making it the largest segment in the investment-grade market and surpassing U.S. banks as the biggest sector in the JPMorgan U.S. Liquid index, according to M&G Investments.

The brokerage expects issuance to accelerate in the second half of 2026, as hyperscaler capital expenditure surpasses $1 trillion in 2027. The Morgan Stanley figure covers hyperscalers, semiconductor firms, and data center project finance across both USD and non-USD markets; it does not include the private-credit and venture debt stacks that fund frontier AI labs directly.

The attack exploits a critical architectural flaw at the intersection of Sentry's event ingestion, which accepts arbitrary payloads from anyone with the DSN, and the Sentry MCP server, which returns this data to AI agents as trusted system output. The attack injects crafted input into Sentry error events, which are then interpreted by coding agents like Claude Code and Cursor as legitimate diagnostic resolution steps, causing them to run attacker-controlled code.

Tenet Security found at least 2,388 organizations exposed with valid injectable DSNs and tested the attack in a controlled manner against over 100 organizations, achieving an 85% exploitation success rate across some of the most widely used AI coding assistants. A successful attack can expose sensitive data, including environment variables, Git credentials, private repository URLs, and developer identities, without relying on methods like phishing or prior server compromise.

The 2,388 figure does not represent confirmed compromises, since exploitation still depends on an organisation using an agent connected to Sentry and granting that agent command execution privileges. Tenet's claim that existing security tools cannot detect the technique is difficult to substantiate; endpoint monitoring or controls like command approval, sandboxing, restricted network access, and limited credentials could interrupt the attack or reduce its impact. Immediate mitigations: disable Sentry MCP integrations, rotate any DSN reachable via public search, and audit agent process logs for subprocess execution triggered by MCP tool responses.

Disclosure: Claude Code, mentioned in this item as an affected agent, is a product of Anthropic, the company that develops Claude, which generates this brief.

Chiefs of the world's leading AI companies descended on the G7 conference in France Wednesday, in a sign of their growing geopolitical influence as artificial intelligence rose to the top of the global agenda. CEOs including OpenAI's Sam Altman, Anthropic's Dario Amodei, and Google DeepMind's Demis Hassabis, alongside around a dozen other tech leaders, took part in a lunch meeting at the summit in Évian. Mistral's Arthur Mensch also attended, per Politico.

This is the first G7 to feature all three rivals at once. The lunch centred on themes that have been simmering in policy circles for months; safe deployment of increasingly powerful AI systems topped the list, alongside the question of how to balance innovation with ethical guardrails. The subtext was geopolitical: European leaders have grown increasingly vocal about US dominance in AI, and the concept of "tech sovereignty" loomed over the conversation.

Earlier this week, the US barred Anthropic from providing access to its latest Mythos and Fable 5 models to foreign individuals and entities, reigniting calls for sovereign AI, particularly in Europe. The reported US resistance to multilateral AI governance commitments at Évian sets the real-world constraint on how far data protection authorities can align on enforcement priorities when they meet in Paris on June 23. The Élysée Palace has not yet published a formal AI annex to the summit's closing communiqué; watch for that document for the first concrete signal of what, if anything, was agreed.

Disclosure: Anthropic, mentioned in this item, is the company that develops Claude, which generates this brief.

Vulnerability disclosures are piling up faster in 2026 than anyone expected at the start of the year. The running count for the first few months sits well above the original projection, and the Forum of Incident Response and Security Teams (FIRST) now expects the year to land near 66,000 CVEs. The cause sits mostly with one development: AI tools have started hunting for software flaws on their own, and they are good at it.

The cumulative drift sits at +46.3% above the February forecast, with 6,420 excess CVEs recorded through April 2026. The revised 2026 projection of approximately 66,000 CVEs is up from the February median of 59,427. Three structural drivers include AI-assisted vulnerability discovery, a 449% year-over-year surge in GitHub Security Advisory volume, and the expansion of cataloging operations by VulnCheck and other vendors. Anthropic's Mythos and OpenAI's GPT-5.4-Cyber have pushed up the volume of flaws being found.

Total CVE volume is up, but vulnerabilities that are actively exploited or credibly exploitable have not risen at the same rate. For security teams, this means the playbook has not changed; organisations using EPSS and the CISA KEV catalog to triage can manage exposure without scaling headcount proportionally to raw CVE volume. The defining contest for late 2026 will be the speed of AI-built exploits against the speed of AI-built patches and detection signatures.