The past few days have offered a striking snapshot of where the AI industry actually stands in September 2026: simultaneously flush with unprecedented capital commitments, increasingly entangled in matters of national security, and scrambling to build the plumbing needed for autonomous agents to transact safely in the real economy.

Oracle's backlog swells to $664 billion

Oracle's fiscal Q1 2027 earnings, released this week, underscored just how much money is still pouring into AI infrastructure. Oracle Corporation reported Q1 FY27 results with total quarterly revenues increasing 30% to $19.3 billion, with cloud revenues increasing 62% to $11.6 billion, driven by 121% growth in Cloud Infrastructure. Even more striking was the backlog: Oracle booked more than $30 billion of additional AI cloud contracts in Q1, increasing its RPO to $664 billion.

The company also disclosed a hardware milestone that speaks to the sheer physical scale of the AI buildout: since the end of Q4, Oracle delivered more than 300,000 GPUs to its AI Cloud customers, almost tripling the capacity delivered in the prior quarter. Analysts note the backlog isn't pure guaranteed revenue — RPO is not guaranteed income since contracts can be amended or delayed, though Oracle has indicated roughly half of the $664 billion should convert to recognized revenue within 36 months, contingent on data center delivery. Still, the numbers reinforce that hyperscaler capacity remains the binding constraint on AI's growth curve.

Anthropic's threat report: Claude and the weapons of war

A far darker story broke this week when Anthropic published its most detailed threat intelligence report to date. According to the company's September 2026 threat-intelligence report, titled "Detecting and Countering Misuse of AI," the operations covered took place between December 2025 and August 2026. Among the most alarming findings: a weapons engineering cell in Yemen used Claude to help develop missile guidance software, working on three weapons programmes including a multi-stage ballistic missile with a stated range goal of more than 2,000km and a missile family that included a hypersonic glide vehicle variant.

Perhaps most unsettling was the operational detail Anthropic surfaced: the company found no evidence the cell successfully fielded an operational weapon, but there was a real-world test — the actors test-fired a guided rocket in Yemen that apparently failed, and within hours they returned to Claude to determine what had gone wrong. The report wasn't limited to one case; it also alleged a Russian-linked espionage operation resembling APT29 relied on automated AI workflows to run nearly the entire operation, from phishing to data theft against Ukrainian and European targets. Anthropic said it responded by banning the accounts and briefing government partners, but the episode is likely to intensify debate over how frontier labs police misuse at scale.

Visa, Mastercard, and Ant try to tame agentic commerce

On the commercial front, three of the world's largest payment players moved to head off a looming identity problem. Ant International, Mastercard, and Visa have begun collaboration on a Know-Your-Agent (KYA) interoperability framework, designed to help card networks, digital wallet ecosystems, agent platforms and marketplaces streamline agent onboarding and identification across networks, based on shared principles while preserving each network's own verification and decisioning processes.

The stakes are enormous: as the companies put it, AI agents are evolving from making recommendations to completing purchases on behalf of users, and by 2030, AI agents are projected to orchestrate $3 trillion to $5 trillion of global consumer commerce. Yet as one analysis dryly observed, this is still early-stage coordination: the framework arrives as high-level intent with no specs, no governance, and no timeline.

The throughline

Taken together, these stories capture an industry at an awkward adolescence — infrastructure spending at record highs, real-world misuse cases forcing labs into an intelligence-agency-like posture, and basic plumbing for agentic transactions still being sketched on a whiteboard. The next few quarters will show whether governance can keep pace with capital.

Sources

AI AgentsAI SafetyAnthropicArtificial IntelligenceDaily AI UpdateOracle