Salesforce acquired Fin, an AI-powered customer service platform, to strengthen Agentforce, its enterprise platform for building custom AI agents. This acquisition represents a major bet by Salesforce to consolidate its enterprise automation offering through intelligent agents.

OpenAI announced a Partner Network program with $150 million in investments to accelerate enterprise AI adoption, deployment, and transformation globally. This initiative reflects OpenAI's strategy to expand its ecosystem beyond direct products.

Indian startup Sarvam achieved unicorn status with a $234 million funding round led by HCLTech, which invested $150 million. This milestone reflects accelerated growth in India's AI ecosystem.

NewCore emerged with $66 million in funding, arguing that managing AI agents as entities with distinct identities will be the next enterprise security challenge. The startup addresses a critical aspect as AI agents integrate deeper into business operations.

Google announced a $1.5 billion investment for 2026-2027 to expand its data center campus in Jackson County, Alabama. The investment reflects the race for computational infrastructure needed to train and serve AI models.

Dozens of cybersecurity experts urged the White House to remove export restrictions on Anthropic's Fable and Mythos models, arguing they would limit defensive cybersecurity capabilities. The debate highlights tensions between national security and enterprise defensive capabilities.

According to Semafor, the White House's decision to impose export restrictions on Mythos was driven by fears that a China-linked group accessed the model. The report intensifies national security concerns around advanced AI models.

Casey Harrell, diagnosed with ALS and paralyzed, has become the "first power user" of a brain-computer interface after almost three years, accumulating thousands of hours of use. The case demonstrates the transformative potential of BCIs in restoring communication abilities.

In April, an Earth observation satellite found what it was looking for autonomously using AI for the first time. This milestone opens possibilities for autonomous environmental monitoring, disaster response, and security applications.

Meta worked with Rank One Computing, whose board includes a former CIA deputy director and former FBI science chief, to develop face recognition for smart glasses. The collaboration raises questions about the relationship between commercial and government surveillance technology.

Researchers found that using LLMs as judges to evaluate models is unreliable: pairwise preferences flip 13.6% between identical evaluations and cross-judge agreement is only 76%. The finding questions the validity of LLM-based leaderboards and evaluations.

Researchers found that LLMs have distinctive, predictable preferences for certain names that vary by model and version, causing specific names to appear recurrently in generated content. The finding helps identify AI-generated content and reveals bias behaviors in models.

Researchers presented Orchestra-o1, a framework enabling efficient coordination of multimodal AI agents for complex tasks involving text, image, audio, and video. The work advances multi-agent systems by solving heterogeneous task decomposition.

HOTE, a framework using hybrid reinforcement learning, was proposed to autonomously evolve agents performing deep research in open-ended tasks. The approach aims to close the gap between static agents and adaptive capabilities for real-world research.

Researchers introduced WebDecept, a framework to test autonomous web agents under realistic e-commerce deceptive interfaces. The study found current agents are highly susceptible to manipulations and prompt-based constraints are insufficient.