Akamai Launches AI Products, Offers Access to Nvidia, Google Models

Key Takeaways

• <ul><li>Akamai launched Inference and Agent Sandbox products for its cloud AI platform</li><li>Akamai Inference provides access to 12 open-weight models including Nvidia Nemotron 3 Ultra and Google Gemma 4</li><li>Akamai Agent Sandbox uses WebAssembly isolation for running AI agents</li><li>Akamai's cloud platform has multiyear contracts worth over $14 billion</li><li>Poisoning just 3% of training data can cause an 11% drop in AI performance</li><li>UT Austin unveiled UT Dell Medicine, an AI-native academic health system</li><li>Zimbabwe launched its first rural AI hub in Gokwe with UN support</li><li>AI in clinical trials can reduce data aggregation time by 90% but adoption remains slow</li><li>UC Berkeley professor Borcherds stopped graded homework due to AI cheating</li><li>Publishers are experimenting with content designed specifically for AI agents</li></ul>

Akamai launches AI inference and agent sandbox products

Akamai Technologies announced two new products for its cloud AI platform. Akamai Inference gives access to 12 open-weight models through OpenAI-compatible APIs. Akamai Agent Sandbox lets developers run AI agents in isolated environments using WebAssembly. The company expanded its GPU fleet to include NVIDIA Blackwell Ultra clusters and is preparing for the Vera Rubin platform in 2027. Both products are available in beta.

Akamai launches AI inference and agent sandbox products

Akamai Technologies introduced Akamai Inference and Akamai Agent Sandbox on October 27, 2026. The inference service provides access to 12 open-weight models including Nvidia Nemotron 3 Ultra and Google Gemma 4. The sandbox uses WebAssembly isolation and is delivered as an open source component of the Spin framework. Akamai's cloud platform has secured multiyear contracts worth over $14 billion and operates more than 4,350 edge locations globally.

AI secretly being embedded across Home Office, report claims

A report by Open Rights Group says AI is being used across the UK Home Office in near-total secrecy. AI tools are used for border surveillance, determining asylum seekers' ages, and prioritizing immigration cases through the Ipic tool. The report says AI can replicate bias and discrimination, especially when driven by anti-immigration policies. Open Rights Group is calling for regulation, transparency, and a public register of government AI tools.

The worst thing about AI music is not the slop

This opinion piece argues that the main problem with AI music is not the quality but that listeners do not understand what they are hearing. AI music can generate complex melodies but often lacks emotional depth and nuance compared to human musicians. The author says this creates a musical flatness that makes it hard to connect with the music. The article calls for a more nuanced understanding of how AI music can expand human creativity.

UT Austin unveils UT Dell Medicine as it bets big on new AI-native hospital

The University of Texas at Austin unveiled UT Dell Medicine, a unified academic health system. It brings together clinics, Dell Medical School, and the upcoming UT Dell Medical Center under one umbrella. The system is being built from the ground up with a focus on artificial intelligence. The goal is to revolutionize healthcare delivery and education through AI-native approaches.

The I'm So Sorry trend uses AI edits then reveals real photos

A viral trend on TikTok and Instagram uses AI to make old photos look flawless, then gradually undoes the edits to reveal the original pictures. The trend aims to remind people that human imperfections are more beautiful than AI-perfect images. Users share candid moments they once considered embarrassing but now laugh about together. The movement celebrates authentic, unedited photos taken before AI image editing existed.

Department of War addresses AI data poisoning threats

The Department of War is working to secure AI systems used in defense operations against data manipulation attacks. Adversaries can compromise AI models by posting as few as 250 malicious documents or by embedding harmful prompts. Poisoning just 3 percent of training data can cause an 11 percent drop in AI performance. CISA's 2025 CSI provides steps to safeguard data from poisoning attacks, including provenance verification and quality controls.

Zimbabwe launches first rural AI hub in Gokwe

Zimbabwe opened its first rural AI, Digital and Innovation Hub at Nyamuroro Vocational Training Centre in Gokwe. The hub was launched by the government in partnership with the United Nations Development Programme and Uncommon.org. It aims to teach young people coding, digital marketing, graphic design, data management and online entrepreneurship skills. Zimbabwe also launched a National Artificial Intelligence Strategy running from 2026 to 2030 to support modernization and economic growth.

Regulatory gaps slow AI adoption in clinical trials

AI in clinical trials can reduce data aggregation time by 90 percent and cycle times by 25 percent, yet adoption remains slow. The main barriers are regulatory uncertainty and organizational bottlenecks rather than technology limitations. Clinical data teams often resist retraining or reorganizing because trial enrollment deadlines leave no time for change. The key factor will be whether the FDA finalizes AI guidance with binding validation requirements.

Publishers experiment with content made for AI agents

Some publishers are creating content specifically designed for AI agents instead of human readers. They hope people will pay for content tailored to AI consumption. However, the business model remains uncertain as publishers question whether humans will pay for content that AI agents could provide for free. The industry is taking a wait-and-see approach to this emerging trend.

UC Berkeley professor stops graded homework due to AI cheating

A UC Berkeley professor named Borcherds stopped giving graded homework because students use AI to cheat. He said online exams are now useless since AI makes scores rise and he cannot tell who is honest. UC Berkeley tried AI-detection software but found it unreliable. The university now lets professors set their own AI rules if they write them in course syllabi. Borcherds believes in-person exams are the best solution left.

Sources

• Akamai launches AI inference and agent sandbox products
• Akamai launches AI inference and agent sandbox products
• ‘Digital hostile environment’: AI secretly being embedded across Home Office, report claims
• The worst thing about AI music is not the slop
• UT Austin unveils UT Dell Medicine as it bets big on new AI-native hospital
• Using AI to make fun of AI: The viral "I'm so sorry" trend that makes reality the punchline
• Securing Defense AI Against Data Manipulation
• Zimbabwe's first rural AI hub set to reduce digital divide
• AI Adoption in Clinical Trials Slowed by Regulatory Gaps, Not Technology Limits
• AI Agents Are Media’s New Favorite Audience
• Fields Medal-winning UC Berkeley professor ends graded homework due to AI cheating — says in-person exams are the only solution left to widespread AI use