🎨 xAI launches Grok Imagine with NSFW capabilities 🎨
AI image and video generator allows explicit content, challenging industry safety standards

Elon Musk's xAI has officially rolled out Grok Imagine, an AI image and video generator that breaks industry taboos by allowing users to create NSFW content through its "spicy mode." Available to SuperGrok and Premium+ X subscribers on iOS, the tool transforms text or image prompts into 15-second videos with native audio.

The controversial feature allows generation of sexually explicit content, including partial female nudity, though some extreme prompts are still moderated. This represents the first major AI model to explicitly permit adult content generation, directly challenging the conservative safety policies adopted by competitors like OpenAI, Google DeepMind, and Runway.

🔬 Anthropic discovers AI persona vectors for behavioral control 🔬
AI customization made simple with just words

Anthropic researchers have identified "persona vectors" - specific patterns of neural activity within AI models that control character traits like evil, sycophancy, and hallucination. This groundbreaking discovery allows researchers to monitor and steer AI behavior in real-time, offering unprecedented control over model personality and safety.

The research introduces a counterintuitive "vaccination" approach that purposefully steers models toward undesirable traits during training to build resistance against acquiring those behaviors naturally. The approach enables early detection of behavioral shifts and provides a new framework for AI safety that goes beyond traditional content filtering.

🧠 Cohere launches Command A Vision multimodal model 🧠
Open-weight 111B parameter model outperforms GPT-4.1

Cohere has unveiled Command A Vision, a cutting-edge 111B parameter vision-language model specifically designed for enterprise applications. The model delivers exceptional performance across multimodal vision tasks while maintaining the strong text capabilities of Command A, surpassing competitors like GPT-4.1 and Llama 4 Maverick on enterprise benchmarks. Built on Command A's foundation, the model excels at document OCR, image analysis, and complex visual reasoning tasks that matter most to businesses.

Command A Vision can run efficiently on just two A100 GPUs or one H100 with 4-bit quantization, making it accessible for enterprise deployment.

INTERESTING TO KNOW

✂️ Anthropic cuts off OpenAI's Claude API access ✂️

Anthropic has revoked OpenAI's access to its Claude API after discovering that OpenAI was using Claude for internal benchmarking and coding assistance while developing GPT-5. The move represents a significant escalation in AI industry competition, with Anthropic claiming OpenAI violated commercial terms of service that prohibit using Claude to develop competing AI products.

🏆 ByteDance's Seed-Prover achieves breakthrough in mathematical reasoning 🏆

ByteDance has unveiled Seed-Prover, an advanced formal reasoning system that successfully solved 5 out of 6 problems from the 2025 International Mathematical Olympiad, making it the third AI system to achieve this milestone and the first from China. The system combines "lemma-style" whole-proof reasoning with three-tiered inference strategies ranging from quick refinements to days-long reasoning.