AI is entering a new phase.
Last week showed how quickly the landscape is shifting:
• AI labs negotiating with governments
• Developers building billion-dollar tools in months
• Frontier models hitting technical limits in long conversations
• New infrastructure for tracking token economics and AI costs
• Companies racing to integrate AI into every interface
At the same time, tensions are rising between technology companies and governments over how AI should be used — especially in defense and intelligence.
Week 10 of Silism breaks down the most important signals behind the headlines.
Need-to-Know
- OpenAI clarified the structure of its Pentagon partnership, highlighting cloud deployment plans and three strict red lines addressing surveillance and weaponization concerns.
- Despite Trump’s restrictions on Anthropic’s Claude, the U.S. reportedly used the model in intelligence analysis and targeting operations related to airstrikes on Iran.
- Researchers from ETH Zurich and Anthropic found that AI models can identify anonymous accounts within minutes with accuracy reaching roughly two-thirds.
- New research shows frontier models such as GPT-5 may lose up to 33% accuracy during long multi-turn conversations, recommending periodic restarts with summarized context.
- Perplexity released two open-source embedding models that rival Google and Alibaba alternatives while requiring significantly less memory.
- Anthropic extended Claude’s memory capability to free users and introduced a migration tool that imports data from ChatGPT and Gemini.
- Cursor reportedly exceeded a $2B annualized revenue run rate, doubling in three months as enterprise adoption surged.
- OpenAI’s Pentagon collaboration sparked backlash, with some users migrating from ChatGPT to Claude while Sam Altman acknowledged the reputational challenge.
- Apple is reportedly evaluating Google infrastructure to power a Gemini-enhanced Siri as it seeks to accelerate its AI roadmap.
- Google upgraded Gemini integration for Google Home, enhancing voice automation and introducing live camera query capabilities.
- Stripe previewed an AI billing system that tracks token usage and automatically applies markup margins for AI-driven SaaS products.
- Meta is testing an AI shopping assistant in the United States to compete with similar tools in ChatGPT and Gemini.
- OpenAI is revising its Pentagon agreement to block intelligence agency usage without additional contractual approvals.
- Meta secured a patent for an AI system capable of simulating social media activity after death or prolonged inactivity, triggering ethical debate around digital afterlife.
- Australia is considering rules requiring app stores to block AI services lacking age verification, with heavy penalties for non-compliance.
- Seven major technology companies signed Trump’s initiative to fund power infrastructure required for expanding AI data centers.
- OpenAI is reportedly building a GitHub-style coding platform that could eventually compete with Microsoft’s developer ecosystem.
- GPT-5.4 is rumored to introduce a one-million-token context window and an “extreme reasoning” mode designed for complex tasks.
- Anthropic is approaching a $20B annual revenue run rate while continuing its dispute with the Pentagon over military restrictions.
- Meta signed a multi-year licensing agreement with News Corp worth up to $50M annually for AI training and content access.
- The Pentagon officially classified Anthropic as a “supply-chain risk,” limiting contractors that rely on Claude from defense work.
- Wired reported that the Pentagon tested OpenAI models through Microsoft Azure OpenAI infrastructure during periods when military usage was officially restricted.
- Netflix is acquiring actor Ben Affleck’s AI film production startup InterPositive and integrating its team, with Affleck joining as a strategic adviser.
- AWS launched Amazon Connect Health, an AI agent platform built to automate administrative processes in healthcare systems.
- Meta will allow third-party AI chatbots on WhatsApp in Europe for 12 months, requiring providers to pay a per-message fee.
- Google added cinematic video overviews to NotebookLM, allowing AI Ultra users to generate animated research summary videos from notes.
Weekly AI Launches
- Notra launched a tool that generates marketing copy and changelogs automatically from GitHub and Linear activity.
- OpenFang released a secure open-source Agent OS built in Rust with sandboxing and scheduling capabilities.
- Voicr introduced an offline privacy-focused voice dictation app that converts speech into customizable text formats.
- Simplora launched a comprehensive AI meeting assistant with advanced preparation workflows and unlimited AI notes.
- GojiberryAI released an intent-based LinkedIn outreach platform that identifies warm prospects and tracks conversion signals.
- Crawler.sh introduced a fast local-first crawler that extracts website content into Markdown while performing SEO analysis.
- NothingHere launched a macOS panic button that hides apps, mutes audio, and opens a fake document instantly.
- WEIR AI introduced an identity protection service that monitors web mentions and enables licensing or defense of personal data.
- Lavalier AI released an AI recruitment assistant that defines roles, generates interview questions, and compares candidates.
- Mailercloud introduced a high-throughput email API platform with real-time tracking and high deliverability for SaaS products.
- Secret Sauce 3D launched an AI co-pilot generating editable geometry, textures, and UV layouts for 3D pipelines.
- Skyvern released MCP integration enabling autonomous AI web automation on newly encountered sites.
- Anything API launched a system that converts browser automations into production-ready APIs for any website.
- Kodo introduced an AI design platform that generates fully editable graphics and presentations from prompts.
- Enia Code launched a proactive coding assistant that detects bugs and suggests architecture improvements while developers type.
- Picsart released Persona & Storyline enabling creators to generate consistent AI characters and narrative videos.
- Aident AI Beta 2 introduced a plain-language automation platform with over 1,000 integrations and centralized monitoring.
- Coursekit launched a no-code AI tutor generator that turns course URLs into 24/7 student learning assistants.
- Heywa released a visual search AI replacing traditional text results with dynamic visual story comparisons.
- Golf launched a governance platform that manages policies, auditing, and security controls for AI agents and MCP servers.
If you want clear insights into where AI infrastructure, regulation, and enterprise adoption are heading — subscribe.
Silism tracks the structural shifts shaping the future of technology.
What caught your attention most Last week?