March 11, 2026 is a busy day: Anthropic creates an interdisciplinary institute for public debate on AI, Perplexity rolls out its “Computer” vision with four simultaneous APIs, OpenAI publishes new agentic primitives in the Responses API, and Meta details four generations of custom AI chips developed over two years.
Anthropic Institute: Jack Clark heads public benefit
March 11, 2026 — Anthropic launches The Anthropic Institute, a new effort to advance public debate on the challenges posed by powerful AI. The initiative is led by co-founder Jack Clark, who takes a new role as Head of Public Benefit at Anthropic.
The Institute will assemble an interdisciplinary team — researchers, economists, lawyers, public policy specialists — with unique access to Anthropic’s cutting-edge models. Its mission: analyze and communicate societal, economic, and security impacts of AI as they evolve.
Three initial hires were announced:
| Hire | Background |
|---|---|
| Matt Botvinick | Resident Fellow, Yale Law School; former Senior Director of Research, Google DeepMind |
| Anton Korinek | Professor of Economics (on sabbatical), Economic Research team |
| Zoë Hitzig | Former OpenAI, specialist in social and economic impacts of AI |
The Institute builds on Anthropic’s existing teams: Frontier Red Team, Societal Impacts, Economic Research. In parallel, Anthropic announces an expansion of its Public Policy team, focused on model transparency, consumer energy protections, export controls, and global AI governance.
The creation of this Institute marks a notable step: Anthropic formally structures its public engagement and gives it a face with one of its co-founders.
🔗 Introducing The Anthropic Institute
Perplexity: coordinated launch of the “Everything is Computer” vision
March 11, 2026 — Perplexity publishes four simultaneous announcements that form a coordinated launch around its “Computer” vision: AI as a personal and professional computer.
Personal Computer and Enterprise
The Personal Computer is a dedicated Mac mini that runs 24/7, connected to local applications and Perplexity servers. It acts as a digital proxy for the user — sensitive actions require explicit approval. Waitlist open.
The Computer for Enterprise connects to Snowflake, Salesforce, HubSpot and hundreds of platforms. Skills are customizable, and Slack integration allows work in DMs or shared channels. It relies on SOC 2 Type II, SAML SSO and audit logs. Perplexity cites an internal-study figure from 16,000 requests: $1.6M saved in labor costs and 3.25 years of work accomplished in 4 weeks.
Comet Enterprise is a native AI browser with admin controls (domain permissions, action logs, MDM) and a CrowdStrike partnership for browser-level protections.
Perplexity Finance receives 40+ live finance tools (SEC filings, FactSet, S&P Global, Coinbase, LSEG, Quartr, Polymarket), with broker connection via Plaid for analysis of a real portfolio.
Premium Sources offers access to Statista, CB Insights and PitchBook directly in research feeds — paid sources are automatically cited.
Agent API: full runtime orchestration
The Agent API is a managed runtime to build agentic workflows with integrated search, tool execution and multi-model orchestration. It replaces in a single integration point a model router, a search layer, an embeddings provider, a sandbox service and a monitoring stack.
| Integrated tool | Capabilities |
|---|---|
web_search | Filtering by domain, recency, date range, language, content budget |
fetch_url | Direct URL retrieval |
| Custom functions | Full support |
Four optimized presets cover use cases: quick factual search, balanced search, deep multi-source analysis, and institutional search. Deep Research 2.0 is available via the advanced-deep-research profile — it launches dozens of searches per query and processes hundreds of documents.
The Agent API is model-agnostic, supports multi-model fallback chains for near-100% availability, and is available today on docs.perplexity.ai.
Search API: improved snippets and SEAL benchmark
The Search API update focuses on snippet quality and evaluation infrastructure.
The new span-level labeling pipeline identifies which segments of a source document are relevant to a query. Result: smaller, more precise snippets, which reduces token costs and improves context management for downstream models.
The SEAL benchmark tests whether a retrieval system can answer questions whose answers change over time. Perplexity advances on SEAL-Hard while other providers decline. The search_evals framework is updated open source on GitHub.
Other improvements: multi-query support (up to 5 in a single API call), filtering by language (ISO 639-1 code) and by country, and a Python SDK (pip install perplexityai) with native support for all three APIs.
Sandbox API: isolated code execution for agents
Perplexity opens its internal code execution environment as a standalone service. Each session runs in an isolated Kubernetes pod, with a mounted persistent filesystem. Supported languages: Python, JavaScript, SQL. Installing packages at runtime is possible.
Sessions are stateful: files created in one step are available to subsequent steps, and long-running workflows can pause and resume hours later. Security relies on a zero-trust model: no direct network access, outbound traffic via proxy, code never has access to raw API keys.
The Sandbox API will be integrable into the Agent API with the same API key and credits. Status: private beta coming soon.
OpenAI: Responses API receives a computing environment for agents
March 11, 2026 — OpenAI publishes an engineering post detailing new primitives in the Responses API to build reliable autonomous agents: a Unix shell tool, hosted containers, native context compaction and reusable agent skills.
The shell tool
The shell tool allows the model to interact with a computer via the command line, with access to classic Unix utilities (grep, curl, awk). Unlike the existing code interpreter that only executed Python, the shell tool supports Go, Java, Node.js and other environments. GPT-5.2 and later models are trained to suggest shell commands.
The Responses API can execute multiple shell commands in parallel via separate container sessions, and enforces an output cap per command to avoid saturating the context window.
Hosted containers
The container forms the model’s workspace:
| Component | Description |
|---|---|
| Filesystem | Upload, organize and manage resources via container and file APIs |
| Databases | Structured storage (SQLite) — the model queries tables instead of loading all content |
| Network access | Centralized egress proxy with allowlist, secret injection by domain |
Native context compaction
For long-running tasks, the Responses API includes a native compaction mechanism: models are trained to produce a compact, encrypted representation of conversational state. Available server-side (configurable threshold) or via an endpoint /compact. Codex uses this mechanism to maintain long-lived coding sessions without degradation.
Agent skills
Skills condition workflow patterns into reusable bundles: a folder with a SKILL.md file and associated resources. The Responses API automatically loads the skill into context before sending the prompt to the model. Skills are manageable via API and versioned.
In parallel, a developer blog post celebrates one year of the Responses API with five customer case studies. The two publications present a coherent picture of the platform’s evolution toward agentic capabilities.
🔗 Responses API + computer environment 🔗 One year of the Responses API
OpenAI: defense strategy against prompt injection
March 11, 2026 — OpenAI publishes a security article on making AI agents resistant to prompt injection attacks.
Early attacks inserted direct instructions into external content (Wikipedia pages, emails). As models improved, attacks evolved into social engineering: convincing professional context, simulated urgency, claimed authorization. A 2025 example in the article showed a successful attack in 50% of cases on an older ChatGPT version.
OpenAI frames the problem as a tripartite system (employer / agent / malicious third party), analogous to a human customer service agent exposed to manipulation attempts. The goal is not to perfectly detect every attack, but to limit the impact of a successful manipulation.
| Countermeasure | Description |
|---|---|
| Source-sink analysis | Detects combinations of untrusted content + dangerous action |
| Safe Url | Detects if conversation information would be sent to a third party — requests confirmation or blocks |
| Application sandbox | Canvas and ChatGPT Apps detect unexpected communications and request consent |
Safe Url also applies to navigations in Atlas as well as searches and navigations in Deep Research.
🔗 Designing AI agents to resist prompt injection
Meta MTIA: four generations of AI chips in two years
March 11, 2026 — Meta publishes a technical article on its family of custom AI chips MTIA (Meta Training and Inference Accelerator). In two years, Meta developed four successive generations to serve billions of users at lower cost.
“AI models are evolving faster than traditional chip development cycles.”
| Generation | Main innovation |
|---|---|
| MTIA 300 | First chip optimized for Ranking & Recommendation models, reusable modular baseline |
| MTIA 400 | Shift toward GenAI workloads, rack of 72 chips in a single scale-up domain |
| MTIA 450 | Doubled HBM bandwidth, +75% MX4 FLOPS, hardware acceleration for attention and FFN |
| MTIA 500 | +50% HBM bandwidth vs MTIA 450, inference-focused GenAI performance |
Progress from MTIA 300 to MTIA 500: HBM bandwidth multiplied by 4.5 and FLOPS multiplied by 25. Meta’s strategy rests on high-velocity development (one new chip per year), an inference focus rather than pre-training, and native PyTorch integration.
“Mainstream GPUs are typically built for the most demanding workload — large-scale GenAI pre-training — while Meta’s primary need is inference.”
The Processing Element architecture combines two RISC-V vector cores, a Dot Product Engine, a Special Function Unit, a reduction engine and a DMA. The software stack relies on PyTorch, vLLM, Triton and dedicated MTIA compilers, with vLLM integration via a plugin architecture.
🔗 Meta MTIA: scale AI chips for billions
Gemini CLI v0.33.0: enriched Plan Mode and A2A authentication
March 11, 2026 — Gemini CLI publishes version v0.33.0, two weeks after v0.32.0 which introduced Plan Mode.
| Category | What’s new |
|---|---|
| Agent architecture | HTTP authentication for remote A2A agents, discovery of authenticated A2A agent maps |
| Plan Mode | Integrated research sub-agents, support for annotations for user feedback, new subcommand copy |
| CLI interface | Compact header with ASCII icon, inverted context window display, 30-day default retention for chat history |
The addition of HTTP authentication for the A2A (Agent-to-Agent) protocol is the main technical novelty: Gemini CLI can now discover and authenticate with remote agents, laying the groundwork for secure multi-agent orchestration.
Gemini in Chrome: expansion to India, New Zealand and Canada
March 11, 2026 — Google expands Chrome’s AI features to three new markets: India, New Zealand and Canada.
Gemini in Chrome — the AI assistant in the side panel, based on Gemini 3.1 — is now available on Mac, Windows and Chromebook Plus in these regions. Deployed features include access to Gmail, Maps, Calendar and YouTube from Chrome, cross-analysis of multiple open tabs, and image transformation directly in the browser via Nano Banana 2. The update adds 50+ additional languages, including Hindi, French and Spanish.
🔗 Chrome expands to India, New Zealand and Canada
--- ## AlphaEvolve: new lower bounds for 5 Ramsey numbers
March 11, 2026 — Pushmeet Kohli (Google DeepMind) announces that AlphaEvolve has established new lower bounds for 5 classical Ramsey numbers in extremal combinatorics — problems so difficult that Erdős himself remarked on their complexity, and whose previous best results dated back at least a decade.
AlphaEvolve acts as a meta-algorithm that automatically discovers the search procedures needed, where historically one had to design specific algorithms by hand. This result illustrates AlphaEvolve’s capabilities beyond the Google kernel optimizations it was already known for.
Gemini Embedding 2: Google’s first fully multimodal embedding
March 10, 2026 — Google announces Gemini Embedding 2, described as “our most capable and first fully multimodal embedding model”. It is Google’s first natively multimodal embedding model, available to developers via Gemini API and AI Studio.
GitHub Copilot: code review from the terminal and JetBrains advances
March 11, 2026 — Two notable updates for GitHub Copilot.
Code Review from GitHub CLI v2.88.0
It is now possible to request a Copilot code review directly from the terminal. The commands gh pr edit --add-reviewer @copilot (non-interactive mode) and gh pr create (interactive mode) integrate Copilot alongside teammates. Reviewer selection benefits from dynamic search, which improves performance in large organizations and fixes accessibility issues. Available on all plans that include Copilot code review — update to GitHub CLI v2.88.0 required.
Agentic improvements for JetBrains IDEs
The JetBrains plugin update brings general availability (GA) for: custom agents, sub-agents and the plan agent, as well as automatic model selection for all plans. In public preview: agent hooks (userPromptSubmitted, preToolUse, postToolUse, errorOccurred) via a hooks.json file in .github/hooks/, and support for AGENTS.md and CLAUDE.md files.
Other improvements complete the release: MCP-configurable auto-approve, a thought panel for long-reasoning models, an indicator for context window usage, and deprecation of Edit mode in the menu.
Explore a repository in Copilot on the web
In public preview: it is possible to explore a repository tree directly from the Copilot web interface. Selected files are automatically added as temporary references to the chat and can be made permanent.
🔗 Explore a repository using Copilot on the web
Anthropic Sydney and Claude for Office
Sydney, fourth Asia-Pacific office
March 10, 2026 — Anthropic announces the upcoming opening of an office in Sydney, its fourth in the Asia-Pacific (after Tokyo, Seoul and Singapore). Australia ranks 4th worldwide for Claude.ai usage per capita; New Zealand ranks 8th by the same metric (Anthropic Economic Index). The office will initially focus on enterprise, startup and research customers.
🔗 Sydney, fourth Asia-Pacific office
Claude for Excel and PowerPoint: shared context and Office Skills
March 11, 2026 — The Claude for Excel and Claude for PowerPoint add-ins receive two major updates: shared context between the two apps (an analyst can extract data from a workbook and use it in a PowerPoint presentation within a single conversation), and the arrival of Skills (one-click reusable workflows) in both add-ins.
Preloaded Skills cover the most frequent use cases: formula auditing, building LBO/DCF models, competitive landscape decks, updating presentations with new data, and investment bank deck review. The add-ins are now available via Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry for compliant deployments. Availability: Mac and Windows on paid plans (Pro, Max, Team, Enterprise).
🔗 Claude for Excel and PowerPoint
NVIDIA Nemotron 3 Super, ComfyUI and GTC 2026
March 11, 2026 — NVIDIA is very active around GTC 2026.
Nemotron 3 Super offers 5× higher throughput for agentic AI compared to the previous generation. It is an open-source MoE (Mixture of Experts) model with 120 billion parameters, optimized for high-frequency inference workloads.
NVIDIA and ComfyUI announced at GDC 2026 (Game Developers Conference) an integration that simplifies local AI video generation for game developers and creators, with support for FLUX and LTX-Video models.
The NVIDIA blog GTC 2026 Live Updates gathers the live announcements from the conference in San Jose — Mistral AI also showcases its frontier models there.
🔗 Nemotron 3 Super 🔗 ComfyUI GDC 🔗 GTC 2026 Live
In brief
Runway Labs — Runway launches an internal incubator led by Alejandro Matamala Ortiz (co-founder, head of innovation). Runway Labs will prototype radically new applications for generative video and General World Models across sectors: film, healthcare, education, gaming, advertising, real estate. Hiring open.
Claude Code /btw — A new command /btw enables side chain conversations while a task is running, without interrupting the ongoing work.
NotebookLM Flashcards — Update to quizzes and flashcards: resume where you left off, track successful or failed flashcards, ability to delete or shuffle flashcards.
Meta Canopy Height Maps v2 — Meta and the World Resources Institute release CHMv2, a new version of global forest canopy height maps. Meta’s DINOv3 self-supervised vision model improves accuracy and global coverage. Applications: climate-driven migration, forest restoration, urban planning. Models available open source.
Z.ai GLM-5 — GLM-5 is now available to Lite users (free tier), after being restricted to Pro users since its February 2026 launch.
What this means
March 11, 2026 illustrates two converging long-term trends.
The first is the platformization of agentic capabilities: OpenAI, Perplexity and GitHub publish complementary primitives on the same day (shell tools, sandboxes, agent hooks, code review). The ecosystem is structuring around reusable blocks — skills, containers, sub-agents — that let developers build reliable agents without reinventing infrastructure.
The second is the race for custom silicon: the detail Meta publishes about its four MTIA generations in two years reveals a clear strategy of independence from consumer GPUs, calibrated for large-scale inference. The same logic drives NVIDIA to release Nemotron 3 Super on the same day as GTC, amid a stream of announcements.
The creation of the Anthropic Institute fits a quieter but lasting movement: as capabilities advance, major AI companies are structuring their public-impact teams — not as PR façades, but as standalone research efforts.
Sources
- The Anthropic Institute
- Sydney, fourth Asia-Pacific office
- Claude for Excel and PowerPoint
- Tweet @bcherny (/btw)
- Everything is Computer — Perplexity
- Agent API — Perplexity
- Search API — Perplexity
- Sandbox API — Perplexity
- Responses API + compute environment — OpenAI
- One year of the Responses API — OpenAI
- Prompt injection defense — OpenAI
- Meta MTIA chips
- Meta Canopy Height Maps v2
- Gemini CLI v0.33.0
- Gemini in Chrome — expansion
- AlphaEvolve — Ramsey numbers
- Gemini Embedding 2
- Copilot Code Review from CLI
- Copilot JetBrains — agentic improvements
- Copilot — explore repository on the web
- Nemotron 3 Super — NVIDIA
- ComfyUI + NVIDIA at GDC
- GTC 2026 Live Updates — NVIDIA
- Introducing Runway Labs
- NotebookLM Flashcards update
- Z.ai GLM-5 Lite rollout
This document was translated from the fr version to the en en language using the gpt-5-mini model. For more information on the translation process, see https://gitlab.com/jls42/ai-powered-markdown-translator