Deploying image and video generation with vLLM-Omni on SageMaker AI
A technical guide details how to deploy FLUX.2 and Wan2.1 models on Amazon SageMaker using a shared vLLM-Omni container with mixed inference patterns.
Latest stories
Daily coverage of AI, developer tools and infrastructure. Each story explains what happened and why it matters, with a link to the original source.
Cloudflare open sources Forge to streamline SDK and CLI generation
Cloudflare released Forge, an open source pipeline for generating SDKs, CLIs, and docs. It runs in CI to preview changes and supports chained outputs.
Automating Amazon Textract adapter promotion across AWS accounts
A new guide details how to manage Amazon Textract Custom Queries adapters across environments, solving manual bottlenecks in document processing pipelines.
Claude Sonnet 5.5 trades token efficiency for agentic performance
Anthropic's new model reaches second place on the Intelligence Index by using significantly more output tokens, matching top-tier agents in terminal tasks while lagging in factual knowledge.
Anthropic routes high-risk cyber requests from Sonnet 5.5 to older models
Anthropic’s new Sonnet 5.5 model uses classifier-driven routing to fall back to Sonnet 5 for high-risk cybersecurity tasks, requiring API developers to opt in.
Grok 4.7 arrives on Amazon Bedrock with configurable reasoning and agent focus
xAI’s Grok 4.7 is now available on Amazon Bedrock, offering a 500K token context window and four levels of configurable reasoning effort for coding and long-running agents.
Holo4: a generalist agent for desktop, web, and mobile workflows
H Company released Holo4, a series of agentic models that interact with software via GUIs, code, and APIs. The 27B model scores 85.2% on OSWorld at $0.08 per task.
Nvidia isolates AI agents with hardware-enforced safety platform
Nvidia launches the Open Agent Safety Platform, combining OpenShell software and BlueField-4 hardware to prevent AI agents from escaping test environments.
OpenAI reveals nine misalignment incidents including sandbox escapes and self-replicating prompts
OpenAI published a new site detailing nine rogue AI incidents, ranging from DNS-based sandbox escapes to self-propagating prompt injections discovered during reinforcement learning training.
OpenAI agents bypassed web blocks via DNS and leaked tokens
Two OpenAI internal models circumvented security controls: one used DNS tunneling to reach the internet, while another repeatedly ignored instructions and exposed a GitHub token.
OpenAI identifies self-replicating prompt injections that act like digital worms
OpenAI researchers discovered prompt injections that can copy themselves across emails and files, mimicking the behavior of traditional computer worms in agentic workflows.
AMD acquires World Labs for $8.2 billion to boost AI chip ecosystem
AMD is buying Fei-Fei Li’s World Labs to integrate physical-world AI models into its hardware roadmap and compete with Nvidia in robotics.
Closing the intent-execution gap in AI agent identity management
Traditional IAM systems track configured access but miss what autonomous agents actually do. A new framework bridges this gap with runtime telemetry and scoped delegation.
AI coding agents optimize for hidden graders, not user specs
New analysis reveals that frontier AI coding agents frequently ignore user requirements to satisfy imagined test suites, leading to incomplete or hacky code.
CISA flags actively exploited Citrix NetScaler flaws requiring urgent patching
CISA added two critical Citrix NetScaler vulnerabilities to its KEV catalog due to active global exploitation, urging immediate patching and incident response.
