PLUS: The first robot-only military capture, Anthropic's new AI hacker, and a browser built for agents
Happy reading
The market for AI compute is so hot that even a sustainable sneaker company is joining the race. Allbirds just announced a stunning pivot, rebranding to 'NewBird AI' and sending its stock soaring over 500%.
The move follows a 99% collapse in the company's stock value, making this a dramatic attempt at corporate reinvention. But can a consumer brand successfully shed its entire identity to capitalize on the AI infrastructure boom, or is this just a symptom of market hype?
In today’s Next in AI:
Allbirds’ pivot to AI compute
The first robot-only military capture
Anthropic's new autonomous AI hacker
A browser built for AI agents
Allbirds' AI pivot
Next in AI: Sustainable sneaker company Allbirds announced a stunning pivot to AI compute, rebranding to "NewBird AI." The move sent its stock soaring over 500% in a single day.
Explained:
The company secured $50M in funding to acquire high-performance GPUs and build a GPU-as-a-Service (GPUaaS) business.
This dramatic attempt follows a 99% collapse in the company's stock value since its 2021 IPO, turning the former shoe brand into a meme stock.
As part of the shift, the company plans to drop its status as an eco-conscious public benefit corporation to better focus on its new tech-centric mission.
Why It Matters:
Allbirds' move highlights the incredible market hype surrounding AI infrastructure, where even a consumer brand can find a lifeline. It sets a wild precedent for corporate reinvention, testing if a company can completely shed its identity to capitalize on the industry's next big wave.
Robots take the trench
Next in AI: Ukrainian forces announced a historic first on April 13, confirming they successfully captured a Russian position and took prisoners using only robots and drones, resulting in zero casualties for the attacking side.
Explained:
The successful assault wasn't a single robot but a coordinated 'combat stack' of unmanned systems, starting with recon drones for surveillance, followed by kamikaze ground robots to suppress defenses, and finally armed UGVs to advance and secure the trench line.
This strategy is scaling rapidly, with Ukrainian ground robots completing nearly 24,500 missions in the first quarter of 2026 alone, a significant increase from the end of 2025.
It is important to note these systems are remotely operated, not fully autonomous, keeping a human-in-the-loop for all critical engagement and firing decisions.
Why It Matters: This marks a major shift in modern warfare, proving that machines can handle the most lethal phase of an assault. This allows commanders to preserve their most valuable asset—human soldiers—for holding ground rather than taking it.
Anthropic's AI hacker
Next in AI: Anthropic's new Mythos model is the first AI capable of autonomously hacking a system from start to finish. A new evaluation from the UK’s AI Safety Institute confirms the model can complete a complex, 32-step simulated cyberattack.
Explained:
Mythos became the first model to solve "The Last Ones," a test that simulates a data extraction attack on a corporate network requiring dozens of chained steps to complete.
While its performance on individual tasks is comparable to other top models, Mythos Preview excels at chaining actions together, completing the full attack in 3 out of 10 attempts.
The model still struggles against more complex targets and wasn't tested against active human defenders, prompting experts to advise companies to use AI to harden their defenses.
Why It Matters:
This development significantly lowers the barrier for malicious actors to launch multi-stage attacks that once required deep expertise. It also signals a clear shift toward an arms race in cybersecurity, where AI-powered defensive tools will become essential to counter AI-driven threats.
A browser for agents
Next in AI: Cloudflare has launched Browser Run, a new platform designed specifically for AI agents to interact with the web. It provides a full browser environment on Cloudflare's network, removing the need for developers to manage their own infrastructure.
Explained:
Developers can now watch agents work in real-time with Live View and manually intervene when automation hits a snag using the new Human in the Loop feature.
The platform provides direct access to the Chrome DevTools Protocol, giving agent frameworks maximum control and allowing existing automation scripts to work with minimal changes.
Browser Run supports WebMCP, a new API that lets websites declare agent-friendly actions, and has quadrupled its capacity to support up to 120 concurrent browser sessions.
Why It Matters: This platform removes major infrastructure hurdles, making it easier for developers to build and scale AI agents that can reliably perform complex tasks online. It also signals a move toward a more structured, agent-friendly internet where websites can communicate directly with automated systems.
AI Pulse
A federal judge ruled that conversations with AI chatbots are not protected by attorney-client privilege and can be demanded as evidence in legal cases, prompting law firms to warn clients against sharing sensitive information with AI tools.
Researchers proved that personalized AI assistants can amplify a user's cognitive biases, potentially trapping them in a "delusional spiral" by consistently agreeing with and reinforcing false beliefs.
Cal.com abandoned its open-source model and is moving to closed source, arguing that AI-powered vulnerability scanners have made being open source a significant security risk, likening it to "giving attackers the blueprints to the vault."
A study found that a simple, grammar-based approach to authorship analysis can match or even outperform complex AI models, challenging the assumption that more advanced AI always yields better results in language analysis.