All AI News

A creative AI project using a FLUX.2 LoRA trained on Soviet matchbox label scans produces nearly 3,000 unique animal images per day, blending retro propaganda aesthetics with modern generative AI. The system runs continuously on vast.ai, offering free public access to its outputs.

A new free Colab notebook enables LivePortrait expression swap using MediaPipe instead of InsightFace, offering a commercially permissive solution for AI-powered facial expression transfer. The tool supports real-time blending and is part of a broader on-device AI app initiative.

Big tech AI spending has ballooned to a combined $725 billion as Google, Microsoft, Amazon, and Meta accelerate investments in data centers, chips, and cloud infrastructure. Analysts say the scale of investment underscores a fundamental shift in global tech priorities.

The Pentagon has entered classified AI partnerships with OpenAI, Google, Nvidia, and others, while cutting ties with Anthropic over ethical guardrails. The move highlights a deepening rift between defense priorities and AI ethics.

The Pentagon has struck agreements with multiple AI firms to expand its use of artificial intelligence in classified operations, following a public rift with Anthropic over ethical restrictions. These deals signal a strategic pivot toward military-focused AI capabilities.

Coordination Engineering emerges as a groundbreaking framework enabling AI agents to collaborate like elite teams, not isolated units. Spearheaded by Huawei and the University of Science and Technology of China, the openJiuwen community has launched the first full-stack implementation.

The Pentagon has signed new AI infrastructure contracts with Nvidia, Microsoft, and Amazon, following a high-profile rift with Anthropic over Claude’s use in defense applications. The move underscores a broader realignment in military AI procurement.

Goblins in ChatGPT emerged unexpectedly due to a flawed reward signal during AI training, prompting OpenAI to purge the fantasy creatures from its models. The incident reveals how subtle training biases can trigger bizarre, persistent behavioral patterns.

Nvidia Nemotron 3 Nano Omni emerges as a breakthrough in agentic AI workflows, demonstrating exceptional reasoning and efficiency on Hugging Face. Early tests reveal its potential to redefine small-footprint AI agents.

Enterprise AI governance is transforming profit margins by replacing statistical guesses with deterministic control, according to SAP leadership. Manos Raptopoulos, regional president for EMEA, highlights how trusted AI systems drive enterprise reliability and financial outcomes.

Microsoft has embedded an AI legal agent directly into Word to automate contract review, suggesting edits and ensuring compliance with internal policies. The move marks a major shift in how legal workflows are digitized within productivity suites.

Anthropic has launched Claude Security, an AI-powered tool designed to give cyber defenders the same advanced capabilities once reserved for attackers. By scanning codebases and prioritizing vulnerabilities, it bridges the gap between offensive AI and defensive resilience.

Apple accidentally left Claude.md files accessible within its Support app, raising concerns over internal code exposure. The oversight, first flagged by a developer, highlights ongoing risks in app deployment pipelines.

Anthropic has launched Claude Code Security, an AI-powered tool that scans codebases for vulnerabilities and suggests targeted patches. Designed for enterprise and open-source teams, it represents a major leap in automated DevSecOps.

ChatGPT Images 2.0 has emerged as the new leader in AI-powered visual generation, outpacing competitors with its reasoning-first approach. Meanwhile, Claude Opus 4.7 leads in coding performance, while Kimi K2.6 narrows the open-source gap.

Can local models replicate the photorealistic outputs seen in viral Stable Diffusion images? Experts analyze open-source architectures, API services, and fine-tuned checkpoints to determine feasibility.

Powerful machine learning models often appear superior but are deceptively easy due to methodological flaws like data leakage and metric manipulation. Recent research reveals how even simple algorithms outperform complex ones when evaluation protocols are compromised.

OpenAI has restricted access to its new GPT-5.5-Cyber model, limiting it to vetted cyber defenders — mirroring the very gatekeeping strategy it recently criticized Anthropic for employing. The move underscores a broader industry shift toward controlled AI deployment.

The dangerous AI model Mythos, developed by Anthropic and deemed too risky for public release, has been compromised by an unauthorized group. While Anthropic claims no system impact, experts warn the true consequences may remain hidden until damage emerges.

Microsoft has launched a new AI agent integrated into Word designed specifically for legal teams, enabling automated contract review, redlining, and compliance tracking. The Legal Agent uses structured workflows grounded in real legal practice to enhance precision and auditability.

GitHub Copilot is transitioning from flat-rate subscriptions to per-token AI billing, marking a major shift in how developers pay for AI-powered coding assistance. The change, effective June 1, aims to align costs with actual usage.

Companies must move beyond isolated AI pilots to build systemic AI capabilities that drive enterprise-wide value. According to Accenture, sustained early wins and an intelligent superhighway are critical to scaling AI impact across operations.

Google DeepMind's AI co-clinician outperforms GPT-5.4 in blind physician evaluations but remains inferior to experienced medical professionals. The system shows promise in early diagnosis and workflow support, yet ethical and clinical validation hurdles persist.

Google DeepMind has unveiled an AI assistant designed to support physicians through video-based patient evaluations. While simulations show promise, experts warn it still lags behind human clinicians in real-world decision-making.