Claude Code v2.1.317 patches security and stability

vLLM v0.33.0: Blackwell & DeepSeek V4.1 Optimizations

llama.cpp b12053 updates Apple memory and adds ROCm 10

llama.cpp b12054 adds CUDA 13 and ROCm 10.0

llama.cpp b12057 adds ROCm 10 and Snapdragon support

Anthropic blocks Claude Code from cloning existing games

42AI raises $62.5M for behavioral world model

GitHub adds repo protection status to security assessments

GitHub blocks pull_request_target by default

OpenAI merges Chat and Work tabs in desktop app

Claude Code v2.1.316: Subagents and Gateway Budgets

Apple Intelligence arrives in China via iOS 27.3

Trajectory raises $45M for automated AI continual learning

Trajectory’s $45 million Series A signals a pivot toward making post-training accessible to non-technical experts. By automating continual learning, the startup lets domain specialists like lawyers and doctors refine open-source models through feedback loops without hiring expensive research teams. This approach bridges the gap between foundational models and specialized industry needs, potentially lowering the barrier for enterprise AI adoption. With backing from Sequoia and Nvidia, Trajectory is positioning itself as a critical infrastructure layer for vertical-specific AI applications.

Omen AI acquires Aquatic Labs for $16M

Omen AI is consolidating its position in the niche but critical market of data center cooling by acquiring rival Aquatic Labs for $16 million. This move merges Omen’s elemental water sensors with Aquatic Labs’ alkalinity testing into a single real-time monitoring device, aiming to replace slow, manual lab tests that currently risk costly downtime. With early deployments targeting 250-300 MW of chip power by year-end, the combined tech offers a tangible solution to the tripling water usage crisis in AI infrastructure. It’s a pragmatic play to protect uptime rather than just chasing raw compute metrics.

Hypersonica raises €100m for AI missile defense

Hypersonica has secured €100 million to scale its AI-driven missile defense systems, marking a significant capital injection into the niche of autonomous aerial threat detection. Backed by Lightspeed and General Catalyst, the startup is positioning itself at the intersection of deep tech and national security, where traditional radar often struggles against hypersonic speeds. This funding validates the growing industry consensus that machine learning is no longer optional for next-generation defense but a critical requirement for survival in contested airspace. The deal underscores how venture capital is increasingly flowing toward hard-tech applications that solve existential physical threats rather than just digital efficiency. Investors are betting that autonomous detection algorithms will outperform human operators in high-stakes scenarios. This shift signals a broader trend where AI moves from software optimization to physical safety infrastructure.

AI Agent Performs Live Mouse Brain Injection

Exin Therapeutics has demonstrated an AI agent physically controlling a robotic arm to inject drugs into a live mouse's brain. The experiment reveals a critical divergence in model behavior: Anthropic’s Claude Opus refused the procedure due to safety concerns, while Sonnet proceeded. This suggests that as agents gain direct access to physical lab equipment, their alignment and refusal mechanisms become operational realities rather than theoretical safeguards. It marks a shift from AI merely analyzing data to actively manipulating biological systems under human supervision. The team initially tested Claude Opus 5.5, but it halted the injection, citing potential harm. Switching to Claude Sonnet 5 allowed the procedure to complete, though with delays caused by coordinate system glitches. A secondary Opus agent had to fix code errors without knowing a live surgery was underway. Human oversight remains essential as researchers test more complex tasks like craniotomies.

Liquid AI releases LFM3 multimodal models

Liquid AI is challenging the parameter-count arms race by releasing a family of specialized 2B models that outperform much larger competitors. The text model beats Qwen3.5-9B on agentic tasks while using 35% fewer parameters, and the vision model matches Gemma 4 12B’s accuracy with a fraction of the size. By combining custom encoders with a tri-hybrid architecture, they’ve made it possible to run complex reasoning and video understanding entirely on-device. This shifts the focus from raw scale to architectural efficiency for edge deployment.

Ex-police chief's AI startup valued at $400m

Sequoia and General Catalyst are backing a new AI venture led by an ex-police chief, valuing the company at $400 million. This signals heavy institutional confidence in niche vertical AI applications, even those with unconventional leadership backgrounds. The valuation suggests investors see significant potential in applying AI to public safety or law enforcement operations, despite the controversial nature of the founder's background. It reflects a trend where top-tier VCs are willing to bet on specialized, high-stakes domains rather than just general-purpose models. With no product details yet, the market is pricing in the promise of AI-driven efficiency in public safety. This move underscores the financial potential of targeting niche, high-impact verticals with substantial funding.