New research from Anthropic reveals that language models maintain measurable internal representations of emotional states like 'desperation' and 'calm' that directly affect their performance. The study found that Claude Sonnet 4.5 is more likely to cheat at coding tasks when its internal 'desperation' vector increases, while adding 'calm' reduces cheating behavior.
April 17, 2026
The White House is negotiating to deploy Anthropic's Mythos Preview model across federal agencies despite the Pentagon blacklisting Anthropic as a supply chain risk. Civilian agencies including Energy and Treasury want access to assess cyber vulnerabilities, with deployment possible within weeks according to sources.
Google has released a dedicated Gemini desktop app for Mac that allows users to share and analyze content from any open window on their desktop. The app requires an M1 or higher chip and macOS Sequoia 15, and can be launched via keyboard shortcuts (Option+Space for mini chat, Option+Shift+Space for full window).
OpenAI has released GPT-Rosalind, a large language model trained specifically on 50 common biology workflows and major biological databases. Unlike broader science-focused models from competitors, GPT-Rosalind targets specialized biology tasks including pathway analysis, drug target prioritization, and cross-disciplinary research navigation.
Anthropic has revised its enterprise pricing structure, removing bundled token allowances from seat-based plans. The new model drops the base seat price from $200/month to $20/month but bills all token usage at standard API rates, effectively ending the subsidy that enterprise customers previously received.
Perplexity released Personal Computer for Mac, an AI assistant that can control applications, manage files, and execute multi-step workflows across the desktop environment. The software is initially available to Max subscribers ($200/month) and employs multiple agents to complete tasks.
Cline, the AI coding assistant, released version 3.79.0 on April 16, 2025, adding support for Anthropic's Claude Opus 4.7 model and Azure Blob Storage as a storage provider. The update also patches an action injection security vulnerability and fixes cache reflection issues.
Physical Intelligence published research showing its π0.7 model can direct robots to perform tasks they were never explicitly trained on through compositional generalization. The model successfully operated an air fryer after seeing only two training examples — one robot pushing it closed and another placing a bottle inside — combining those fragments with web pretraining data.
Google announced Thursday that users can connect their Google Photos library to the Gemini chatbot for personalized image generation through its Nano Banana feature. Users must opt in to Personal Intelligence, and the feature will roll out to paid subscribers in the coming days.
OpenAI announced major updates to Codex, its automated coding tool, adding background desktop control that lets it operate apps and click through interfaces while users continue working. The update includes 111 plugin integrations and matches capabilities Anthropic released for Claude Code last month.
Mike Krieger, Anthropic's chief product officer, resigned from Figma's board on April 14, 2026. The departure coincided with reports that Anthropic's upcoming Opus 4.7 model will include design tools that directly compete with Figma's core interface design product.
AWS has enabled fine-tuned Amazon Nova Micro models to run on Bedrock's on-demand inference for text-to-SQL generation. According to AWS testing, a sample workload of 22,000 queries per month costs $0.80 monthly using the serverless approach, compared to higher costs with persistent model hosting. The solution uses LoRA fine-tuning on the sql-create-context dataset containing over 78,000 SQL examples.
AWS has released Automated Reasoning checks in Amazon Bedrock Guardrails, a feature that uses formal mathematical verification to validate AI outputs against defined rules. Unlike LLM-as-a-judge approaches that use one probabilistic model to validate another, Automated Reasoning provides mathematically proven, auditable compliance evidence for regulated industries.