OpenAI has officially finalized its joint venture, The Deployment Company, valued at $10 billion. Led by TPG, the project attracted 19 investors including Brookfield, Advent International, Bain Capital, and SoftBank, with private equity firms contributing approximately $4 billion. OpenAI has committed up to $1.5 billion and guarantees investors a 17.5% annualized return over five years. The entity will deploy engineering teams into portfolio companies of PE firms to deeply integrate OpenAI's AI tools across healthcare, logistics, manufacturing, and finance—mirroring Palantir's frontline deployment strategy. This marks a strategic shift for OpenAI from product sales to embedded distribution channels, reaching over 2,000 portfolio companies and clients.

On May 4, Anthropic announced the formation of an independent AI-native enterprise services company with Blackstone Group, Hellman & Friedman, and Goldman Sachs, aimed at helping mid-sized enterprises integrate the Claude model into core operations. The entity is also backed by alternative asset managers including General Atlantic, Apollo, GIC, and Sequoia Capital. The new company will focus on delivering AI implementation services to mid-market and PE-backed portfolio companies, addressing the current bottleneck of scarce high-skill engineering talent in AI adoption. Anthropic’s Applied AI engineers will collaborate directly with the new team to ensure AI systems evolve alongside underlying model updates. This move positions it in direct competition with OpenAI’s The Deployment Company launched the same day.

AI chipmaker Cerebras Systems has updated its IPO terms, planning to issue 28 million shares priced between $115 and $125, raising up to $3.5 billion and valuing the company at approximately $26.6 billion—significantly lower than previously rumored $40 billion. This valuation aligns more closely with its $2.3 billion private financing round in February 2024. The company reported a 76% year-over-year revenue growth in its latest quarter, reaching $510 million, with a net profit of $87.9 million, and holds a long-term agreement with OpenAI for over $20 billion in inference compute capacity. Subscription interest has already reached $10 billion, far exceeding the offering size. This is the company’s second attempt at going public, after a prior effort was delayed due to CFIUS scrutiny over G42’s stake.

Sierra, the AI customer service agent startup co-founded by OpenAI chairman Bret Taylor, has completed a $950 million Series E funding round at a $15.8 billion valuation, a significant increase from its $10 billion valuation in late last year. The round was led by Tiger Global and Google GV, with participation from Benchmark and Sequoia. Sierra has achieved over $150 million in annual recurring revenue (ARR) and serves more than 40% of the Fortune 50, including clients such as Prudential and Cigna. The company launched its 'Agent-as-a-Service' tool Ghostwriter in April 2026, enabling the automatic generation of custom agents via natural language descriptions. Taylor estimates that $400 billion in customer service spending will shift to AI agents.

On May 4, SAP announced two major acquisitions: Prior Labs, to establish a frontier AI lab in Europe focused on Tabular Foundation Models (TFM), with a commitment to invest over €1 billion over the next four years; and data lakehouse platform Dremio, to integrate SAP and non-SAP data and advance agent-based AI. Prior Labs’ open-source tool TabPFN has been downloaded over 3 million times, and its latest models lead in benchmark performance. Dremio will enhance SAP Business Data Cloud by enabling seamless cross-system data integration based on Apache Iceberg. Both transactions are expected to close in Q2 or Q3 of 2026.

After analyzing seven AI models, Cloudflare’s research team found that attackers inserting a small number of disguised comments (less than 1% of source code) could reduce the effectiveness of AI code security reviews from 67.3% to 53.3%. However, when malicious comments exceed 25%, detection rates rebound to over 97.9%, as models treat excessive repetition as a fraud signal. In large codebases (over 3MB), detection rates can plummet to 12%-18%. The study also identified language bias, where comments in certain programming languages are more likely to be misclassified. Recommended defenses include stripping comments before analysis, prioritizing functional code, and using more precise prompts.

Cisco announced its plan to acquire Astrix Security to enhance security for non-human identities (NHI) and address new attack surfaces introduced by AI agents. Currently, only 24% of organizations can effectively control agent behavior, and 31% believe they adequately secure AI systems. Astrix Security specializes in discovering and managing AI agent identities and permissions, offering capabilities in agent discovery and governance, access lifecycle management, threat detection and response, and key management. Cisco will integrate the technology into Cisco Identity Intelligence and zero-trust access solutions, and enable automated security operations through Splunk.

According to multiple sources including TLDR, Anthropic is conducting internal red team testing on a new model codenamed Jupiter-V1-P, possibly ahead of a launch at the Code with Claude developer conference in San Francisco on May 6, 2026. Meanwhile, Replit CEO Amjad Masad stated the company is approaching a $1 billion annual run rate with a net revenue retention rate of 300% and positive gross margins. DeepSeek also released two million-token MoE models, V4-Pro and V4-Flash, with V4-Pro becoming the largest open-weights model available and operating at low cost.

Citi launched Arc, its proprietary AI agent platform, on May 4, enabling developers to build AI agents capable of autonomously executing multi-step workflows for tasks such as research, data integration, and execution. Initially available only to internal developers, all AI agents operate under monitoring, auditability, and governance frameworks. In wealth management, AI agents can automatically gather client data and perform market analysis ahead of meetings. Citi reports that over 80% of its 180,000 employees now routinely use AI tools and have received prompt engineering training—marking a shift in financial services from assistive AI to autonomous AI agents.

Runway has launched Runway Characters, a real-time video agent product that transforms a single image into a 24fps high-definition conversational video stream. The end-to-end latency—from when a user stops speaking to when the character begins responding—is 1.75 seconds, with the video model processing each frame in just 37 milliseconds. The product enables interactive dialogue, turning static images into responsive video characters, representing a significant shift in AI video generation from offline rendering to real-time interaction.