TELEPORT

Giving AI agents infra access shouldn't mean security breach

The AI agent ecosystem is moving fast, and that speed creates a real infrastructure problem. Most teams end up with tangled IAM policies, hardcoded secrets, and standing privileges that are difficult to manage and even harder to audit.

Teleport Beams takes a different approach: Every agent runs in its own isolated Firecracker VM with built-in cryptographic identity. Access is precisely scoped and short-lived, without secrets to rotate or privilege creep to manage.

Connections to internal infrastructure and external inference services are handled through identity-based access without the credential overhead that typically comes with multi-agent systems. Every interaction is logged, giving security teams full visibility without additional instrumentation.

A simpler, more secure way to run AI agents in production.

Explore Teleport Beams

_{*This is sponsored}

THROUGH THE VALLEY

1/ Anthropic signs major compute deal with SpaceXAI

SpaceXAI announced a deal that will give Anthropic access to the full capacity of its Colossus 1 data center in Memphis, Tennessee.

The site reportedly delivers more than 300 megawatts of power and over 220,000 NVIDIA GPUs, with Anthropic expected to begin using the compute within the month.

The announcement comes shortly after Elon Musk said xAI would be dissolved into a new entity called “SpaceXAI,” further tying the company’s AI ambitions directly to its infrastructure and space operations.

Anthropic also said the deal will increase limits across Claude Code and the Claude API (including doubling Claude Code’s 5-hour limits for Pro, Max, and Team users). The company’s annualized revenue reportedly surpassed $30B last month, with CEO Dario Amodei saying the business could grow 80x this year.

Why this matters:

The main challenge in AI is increasingly becoming infrastructure, not demand. Companies now need enormous amounts of power, GPUs, and data centers to keep scaling. That’s turning compute into one of the most valuable assets in the industry.

The deal also hints at a larger shift: SpaceXAI is positioning itself less like a traditional AI lab and more like a company focused on supplying large-scale AI infrastructure to others.

2/ The U.S. government wants to inspect frontier AI models before release

Google DeepMind, Microsoft, and xAI agreed to give the U.S. government early access to unreleased AI models for national-security testing.

The evaluations will run through the Commerce Department’s Center for AI Standards and Innovation, which says it has already tested more than 40 models, including systems that were never publicly released.

The agreements are partly driven by concerns around advanced capabilities in areas like cybersecurity and other national-security risks.

Why this matters:

AI companies are no longer building systems first and dealing with regulators later. Governments are now getting visibility into some models before the public does. That’s a sign frontier AI is starting to be treated more like strategic infrastructure than normal consumer software.

3/ Genesis AI unveils robotics model built for human-level hand control

Genesis AI introduced GENE-26.5, a robotics model designed for tasks that require precise hand movements and coordination.

To train the system, the company also built a robotic hand and motion-capture glove that lets humans generate training data simply by performing tasks while wearing the glove. This is a major advancement in robotic reasoning and control.

Genesis says the hardware is around 100x cheaper than traditional systems and can collect data much faster.

The company demonstrated robots cracking eggs, slicing tomatoes, conducting lab experiments, solving a Rubik’s Cube, and playing piano with surprisingly fluid hand control.

Why this matters:

The biggest problem in robotics is increasingly becoming data. Robots need massive amounts of real-world examples showing how humans move and manipulate objects.

Genesis is trying to solve that by turning normal human activity into training data. If that works, robots could improve much faster because every interaction becomes part of the learning process.

TRENDING TOOLS

WHAT I'M CONSUMING

THE VALLEY GEMS

What’s trending on social today:

THAT’S ALL FOR TODAY

Thank you for reading today’s edition. That’s all for today’s issue.

💡 Help me get better and suggest new ideas at barsee@aivalley.ai or @heyBarsee

👍️ New reader? Subscribe here

Thanks for being here.

REACH 100K+ READERS

Acquire new customers and drive revenue by partnering with us

Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.

If you’re interested in sponsoring us, email barsee@aivalley.ai with the subject “AI Valley Ads”.