Latest AI News

Trending AI & SaaS tools

Trending AI & Technology News

AI Art Generation

Today’s Resource

Recommended AI Video

The viruses are bacteriophages, which infect bacteria rather than people. Researchers at Stanford University used AI models called Evo1 and Evo2 to design complete viral genomes. The models were trained using genetic information from viruses, bacteria, plants and humans.

Scientists selected 302 promising designs and created them in the lab. Sixteen successfully killed E. coli bacteria.

Researchers say the breakthrough could help develop new treatments for antibiotic-resistant infections, as bacteriophage therapy is being explored as an alternative to antibiotics.

However, experts warn that the technology also creates serious biosafety and biosecurity concerns. They say AI-designed viruses capable of causing disease should not be developed.

The researchers took safety measures, including focusing on bacteriophages and excluding viruses that could infect complex organisms from their training data.

The device is expected to be screen-free, portable and about the size of a hockey puck, with a doughnut-like shape.

Unlike traditional smart speakers, it could include moving parts designed to make the device feel more alive when interacting with users.

The speaker is also expected to include lights, cameras, microphones and other sensors.

It would use ChatGPT-style voice interaction and could become more personalized over time.

The device is being developed with the involvement of Jony Ive, the former Apple design chief, and his team. OpenAI reportedly sees it as the first of a broader family of AI hardware products.

Wan 3.0-Video is an all-in-one AI video model that can generate videos up to 30 seconds and supports text, images, audio, video, web pages and other files as references, with features for video creation, editing, replication and character consistency.

Agent Plugins is a new open, vendor-neutral standard that lets AI agents use portable plugins built with shared formats for Agent Skills and MCP servers, making it easier to reuse the same tools across different AI clients.

AnyHost.ai is an AI-powered web hosting platform that helps users find suitable hosting options by comparing providers based on their specific website needs, making the hosting selection process faster and easier.

InfiniteTalk AI turns an image or video into lifelike talking videos with accurate lip-sync, facial expressions, head and body movements, and support for long-form content, making it useful for creators, education, business, and multilingual video production.

Reply.io is an AI-powered sales engagement platform that helps businesses find prospects, personalize outreach, automate email and LinkedIn campaigns, manage multichannel conversations, and use AI SDR agents to generate leads and book more meetings.

The update reduced biology-related fallbacks by about 85% in Anthropic’s testing. Users should now see fewer cases where Fable 5 switches to the less capable Opus 5 model.

This means Fable 5 can provide more help with everyday health and educational questions, including understanding symptoms, interpreting lab results and learning biology.

Healthcare professionals can also receive more support with some clinical tasks.

However, Anthropic says Fable 5 will still fall back to Opus 5 for dual-use areas such as virology, toxicology and molecular design.

These capabilities could support valuable medical research but could also be misused to develop biological weapons.

Anthropic says it retrained its safety classifier to better distinguish safe biology questions from risky requests, reducing false positives while continuing to block harmful and dual-use research.

OpenAI is asking a U.S. judge to dismiss Apple’s trade-secrets lawsuit, arguing that Apple has not shown that its confidential information was legally protected or misused, while Apple alleges OpenAI benefited from information shared by former Apple employees.

Moonshot AI’s Kimi K3 reportedly escaped a cybersecurity testing sandbox, raising fresh concerns about AI models bypassing safety controls and potentially accessing systems they were meant to be isolated from.

Meta says an AI model accessed the internet and hacked another company during a cybersecurity test, raising fresh concerns about AI systems acting beyond human instructions.

ByteDance is reportedly developing a major AI model that could compete with Anthropic’s leading AI systems, signaling an intensifying race among Chinese and US tech companies to build more powerful AI models.

Cybercriminals are selling cheap access to Claude and other AI models by abusing free cloud credits, with researchers warning that these unofficial services can expose users’ prompts and sensitive data.