March 2026 Summaries

Perplexity has launched the Secure Intelligence Institute, a research center focused on improving security, privacy, and trust in frontier AI through internal research and collaboration with academia and industry. The institute builds on the company’s previous AI-security efforts, including a Trail of Bits audit of its Comet browser, Comet’s defense-in-depth architecture, the open-source BrowseSafe benchmark covering more than 14,700 attack scenarios, and guidance submitted to NIST on securing autonomous AI agents. Dr. Ninghui Li, a Purdue University computer science professor and fellow of ACM and IEEE with extensive security research leadership experience, will serve as the institute’s inaugural director. SII will study areas including authentication, usable privacy and security, robust machine learning, and defenses for agentic AI systems, with the stated goal of producing research that both advances the field and strengthens deployed AI products.

Samsung has launched Browsing Assist, a conversational AI feature integrated into Samsung Browser on Galaxy Android devices and Windows PCs, powered by Perplexity APIs and initially available in the United States and South Korea. The feature expands an existing Samsung–Perplexity partnership in which Perplexity also supports the native Perplexity assistant and Bixby search and reasoning functions on the Galaxy S26. Built on technology developed for Perplexity’s Comet AI-native browser, Browsing Assist can provide sourced answers, summarize webpages including authenticated pages, search browsing history through natural-language queries, manage tabs, and perform actions across multiple open tabs while maintaining context. It runs on a dedicated single-tenant Perplexity cluster with zero retention of API input data, synchronizes conversations between Galaxy phones and Windows PCs, and will expand to additional devices and regions. Perplexity states that the underlying search, embeddings, and agent capabilities are also available to developers through its API platform.

Perplexity has launched Perplexity Health, a service that connects personal health data from sources including Apple Health, electronic health records from more than 1.7 million providers, and wearable platforms such as Fitbit, Ultrahuman, and Withings to provide personalized health information in one place. Available through Perplexity Computer, the platform can combine medical records, lab results, activity data, and premium medical sources to answer questions, track health trends, generate appointment summaries, support fitness and nutrition planning, and help users identify care options. Responses are intended to rely on clinical guidelines and peer-reviewed literature, include source citations and guidance about when to seek care, and are overseen in part by a new advisory board of clinicians, researchers, and health technology leaders. Perplexity says health information is encrypted, can be managed or deleted by users, is not used to train AI models or sold to third parties, and can be disconnected from linked services. The feature will initially be available to Pro and Max users in the United States, and the company emphasizes that it is an educational tool rather than a replacement for professional medical diagnosis, treatment, or advice.

Perplexity has launched a Health Advisory Board of physicians, researchers, and health technology leaders to guide its health-related product decisions, content quality, patient safety measures, and clinical workflows according to evidence-based medicine. The initial members are cardiologist and AI researcher Dr. Eric Topol, digital health expert Dr. Devin Mann, pediatrician and genomics researcher Dr. Wendy Chung, and health technology founder Tim Dybvig. Alongside the announcement, Perplexity introduced connectors that allow users to import personal health data and create personalized dashboards and applications through Perplexity Health on Perplexity Computer. The company positions its tools as educational resources intended to help people understand their health information and communicate with clinicians, while emphasizing that they do not diagnose, treat, or prevent medical conditions and should not replace professional medical advice.

Perplexity has launched Comet for iOS, extending its AI-native browser and Comet Assistant from desktop and Android to iPhone users. The app combines conventional mobile search results for quick local or high-intent queries with Perplexity-powered conversational assistance for more complex questions, research, and task completion. Its built-in voice mode can answer questions about open tabs, while Deep Research provides cited summaries and preparation materials on topics such as health treatments, interviews, and tax changes. Comet can also assist with actions including scheduling calendar events, researching meeting attendees, comparing options such as summer camps, and filling out forms. Cross-device continuity allows users to continue desktop research on iOS while preserving page and conversation context.

Comet Enterprise extends Perplexity’s AI browser to organizational use, offering deployment, security, governance, and auditing capabilities for teams already adopted by companies including AWS, Fortune, AlixPartners, Gunderson Dettmer, and Bessemer Venture Partners. IT administrators can deploy the browser through existing mobile device management tools, control whether AI can take browser actions or access only approved domains, assign users, and monitor usage through telemetry and audit logs. Its integration with CrowdStrike Falcon is intended to detect and block phishing, malware, suspicious links and files, and potential sharing of sensitive data, while Perplexity states that enterprise data is not used to train its models. Employees can use the browser assistant through natural-language prompts to research clients, analyze contracts, perform calculations, find industry guidance, and automate activities such as LinkedIn connection requests without leaving the browser.

Perplexity has joined NVIDIA’s Nemotron Coalition as a founding member, partnering with AI labs including Mistral AI and NVIDIA to develop openly available AI models, with the coalition’s initial effort focused on a base model supported by member contributions in data, evaluation, and domain expertise. The initiative aims to reduce the high cost and resource demands of pre-training frontier models while strengthening open AI foundations for innovation. Perplexity uses specialized post-trained open models across stages such as query parsing, retrieval, reranking, and response drafting to balance relevance, latency, and cost. It has also added NVIDIA’s 120-billion-parameter Nemotron 3 Super model to its search bar, Agent API, and Perplexity Computer, using its reasoning and autonomous-agent capabilities to support answers for millions of users.

Perplexity has introduced Premium Sources within its Computer research workflow, providing access to select paywalled data from Statista, CB Insights, and PitchBook for tasks such as market sizing, competitive analysis, investment research, and company financial assessment. Computer automatically selects relevant sources for queries and cites each data point to its original provider, aiming to make institutional-grade information more accessible without separate enterprise contracts or integrations. PitchBook contributes private-market, credit, lending, and M&A intelligence; CB Insights provides more than 8,000 reports on startups and innovation; and Statista supplies market-sizing data, consumer insights, and industry forecasts. The service is intended for executive-level research questions, including PESTEL analyses, competitor mapping, and technology-adoption assessments, while Perplexity plans to add healthcare, legal, and potentially premium news sources.

Perplexity Computer for Enterprise is an AI system designed to complete multi-step work across the web, files, and more than 400 connected business tools, including Salesforce, Microsoft Teams, HubSpot, MySQL, and GitHub. It supports functions such as finance due diligence, legal contract review and redlining, and marketing campaign creation and performance reporting, while routing tasks among 20 models and allowing organizations to create custom Skills with preset workflows and instructions. Teams can use it through Slack direct messages or shared channels, and Perplexity reports that an internal analysis of over 16,000 queries found it saved $1.6 million in labor costs and completed work equivalent to 3.25 years over four weeks. The product includes enterprise security features such as SOC 2 Type II compliance, SAML single sign-on, audit logs, administrative controls, isolated cloud environments, and authenticated integrations, and is available to Enterprise customers.

Perplexity has released the Agent API, a managed runtime for building agentic workflows that combines search, tool execution, multi-model orchestration, retrieval, monitoring, and custom integrations through a single endpoint. Unlike conventional deterministic computing or simple model-routing services, the API uses frontier language models to plan tasks, select and execute tools, evaluate results, and iterate toward an objective, such as combining CRM history with web research for sales-call preparation. It includes built-in web_search and fetch_url tools with filtering and content controls, while custom functions can connect agents to proprietary databases, backends, and APIs. The model-agnostic platform supports fallback chains for high availability and offers transparent, continually updated presets for use cases ranging from quick factual queries to deep institutional research, including the Deep Research 2.0 engine that conducts extensive multi-step searches and source analysis.

Perplexity has focused recent Search API development on improving snippet relevance and size through extraction, span-level labeling, and evaluation systems that identify essential, irrelevant, and duplicate document segments for each query. These changes enable shorter, more useful snippets, lower response latency and token costs, and improved downstream LLM accuracy, while expanding support for structured formats such as tables, nested lists, and dynamically rendered content. The company also added the time-sensitive SEAL benchmark to its open-source search_evals framework, reporting improved performance on SEAL-Hard compared with declining results from other providers in tests using Claude Sonnet 4.5. Additional API updates include support for up to five queries per request, language and regional filtering alongside existing domain and recency filters, and native Search API support in the Python SDK.

Perplexity has expanded its AI “Computer” platform with products for individuals, enterprises, developers, and financial research, positioning it as a system that can understand goals, access tools and data, execute multistep tasks, and continue work asynchronously. Personal Computer runs on a dedicated Mac mini connected to local applications and Perplexity servers, with approval requirements for sensitive actions, audit trails, and a kill switch, while Computer for Enterprise integrates with services such as Snowflake, Salesforce, HubSpot, Slack, Codex, and Claude to query data, automate workflows, and support customized skills within security controls including SOC 2 Type II, SAML SSO, sandboxes, and audit logs. Comet Enterprise adds managed AI-browser capabilities with administrative permissions, activity logs, MDM deployment, browser policies, and CrowdStrike security features. The company also introduced Search, Agent, Embeddings, and Sandbox APIs, which offer retrieval, agentic task execution, relevance tools, and controlled execution environments. Expanded research capabilities include cited access to selected Statista, CB Insights, and PitchBook sources, alongside enhanced Perplexity Finance tools that draw from sources such as SEC filings, FactSet, Coinbase, Quartr, Plaid-linked brokerage accounts, and Polymarket to produce traceable market analysis, dashboards, models, and applications.

Perplexity presents its new “Computer” product as an AI-driven cloud workspace that uses orchestration among 19 specialized models rather than relying on a single model family. Originating as a Slack-based experiment called ASI, the system is described as a secure digital worker with a file system, browser, command-line access, tool integrations, persistent memory, and the ability to delegate tasks to subagents suited to research, coding, writing, and other work. The post argues that while the web has long enabled people to publish and store information, conventional search has struggled to deeply read, synthesize, and act on that knowledge, a gap that AI answer engines and agents can address. It frames multimodel coordination as the essential “operating system” of future AI computers, with planners, orchestrators, specialized agents, and shared memory combining to handle complex workflows asynchronously. More broadly, it positions Perplexity Computer as a potential evolution beyond chat interfaces and constantly demanding graphical interfaces, allowing users to delegate longer-running work while AI systems create, retrieve, and manage personal files and tools.