Member of Technical Staff - Secure Intelligence Institute
About the job
- Hold a PhD or have equivalent research experience in Computer Science, Computer Engineering, or a related field, with a primary focus on security and/or privacy.
- Have experience publishing original, impactful research at top security conferences, including IEEE S&P, USENIX Security, ACM CCS, or NDSS.
- Have deep expertise in one or more of the following: security of agentic systems, systems security, web and applications security, program analysis, or software security.
- Demonstrate proficiency in Python.
- Operate with high independence, take ownership, and be comfortable in a fast-paced environment where research directly informs product.
- Communicate clearly and concisely, translating complex attack narratives into actionable insights for engineering and leadership.
- Develop threat models for emerging attack surfaces in AI-native products, including browser, search, and autonomous agents.
- Identify and analyze security and privacy threats across AI systems, infrastructure, and user-facing products.
- Develop novel defenses, mitigations, and detection mechanisms for security and privacy in AI-native products.
- Build security evaluation frameworks, benchmarks, and datasets to measure the effectiveness of different defense mechanisms.
- Partner with the Security Engineering team to translate state-of-the-art research into shipped security features and hardened system architectures.
- Collaborate with academic and industry researchers in the external research network.
- Publish findings at premier venues and contribute to the broader security research community.
- TypeScript, Go, and/or Rust proficiency.
About the company
Perplexity AI builds AI search and research tools for individuals and organizations, combining web retrieval with written answers that link to their sources. Users ask questions in its website or apps, upload files for analysis, and follow up in a conversation, while its Comet browser brings an AI assistant into everyday browsing. Enterprise accounts connect internal files and applications with web research, and developers use its search and agent APIs to add these capabilities to their own software. The company earns money from paid personal and business subscriptions and usage charges for its APIs, alongside free consumer access. Perplexity AI was founded in 2022 and is headquartered in San Francisco, California.
Company Size
201-500
Company Stage
Private
Total Funding
$1.8B
Headquarters
San Francisco, California
Founded
2022
Get referred to Perplexity AI
See people who can refer or advise you
Simplify's Take
What believers are saying
- Business adoption extends beyond consumers: Perplexity reports serving more than 50,000 organizations through Enterprise.
- Distribution reaches business meetings: Zoom's AI Companion embeds Perplexity's API for real-time web answers.
- Employees get a coordinated break: Perplexity's US benefits include a shutdown between Christmas and New Year.
What critics are saying
- Cloud commitments are large: January 2026 reporting valued Perplexity's three-year Microsoft agreement at $750 million.
- Office time is substantial: Perplexity's in-person teams work four days a week on site.
- Platform owners retain control: August 2026 appeals ruling preserved Amazon's ability to restrict users through contractual terms.
What makes Perplexity AI unique
- Orchestration spans model vendors: Perplexity Computer puts 20 advanced models to work in one system.
- Search becomes a reusable component: Perplexity's Search API returns ranked results with domain filters and content extraction.
- Local tasks avoid metered credits: Perplexity Portable Computer runs on supported AMD PCs without using Computer credits.
Help us improve and share your feedback! Did you find this helpful?
Benefits
Health Insurance
Dental Insurance
Vision Insurance
401(k) Retirement Plan
Company Equity
Company News
New study ranks how well AI tools help shoppers hunt for deals. By Nydia Han, Cheryl Mettendorf Friday, October 9, 2026 2:20PM As more shoppers turn to artificial intelligence tools to help find deals, a recent study urges consumers to verify product information and pricing, as some platforms appear to be more reliable than others. About 40% of consumers are using AI for shopping, according to Product AI. According to Dakota Nunley of Product AI, the company asked the platforms 220 shopping questions on everything from pricing to comparisons of products across nine different product categories and recorded those results. Their study found numerous errors across the platforms. "Everything from MacBook pricing being off by $300 in some cases, to SPF sunscreen being off by 10 points, to earbud battery life being off by six hours," Nunley said. Despite the mistakes, the study found differences in performance among the platforms. "But some tools are better than others?" asked Nydia Han. "That's what we saw," Nunley responded. Nunley noted that AI models are constantly changing and improving, meaning results from studies such as this one can vary over time. Among the platforms tested, Product AI ranked Gemini last. Claude finished third, followed by ChatGPT in second place. "Gemini, over half of the questions and shopping-related questions that we asked did have some sort of costly error for consumers," Nunley said, adding that the finding applied "on both the free and paid tiers." Perplexity was identified as the top performer. Regarding Perplexity's performance, Nunley said, "In our findings, we didn't find that perplexity. At this moment was the best choice for consumers." Google, which owns Gemini, disputed the study's relevance to consumers, telling Action News that Product AI did not evaluate the Gemini consumer app and instead used Gemini software designed for developers. According to Google, "The Gemini app includes everyday user features... including shopping experiences powered by billions of continually-refreshed product listings." Users should also note that Perplexity differs from other platforms because it uses AI to actively search the internet and other AI models before providing information to consumers. For shoppers looking for products on Amazon, keep in mind that Amazon uses AI shopping assistant, Rufus, for Amazon product and pricing information. Amazon has blocked other AI bots from searching its marketplace. TikTok also recently launched its own AI shopping tool, adding to a growing number of AI-powered shopping options available to consumers.
Perplexity AI releases pplx-embed-v2-late: A 0.6B edge model and a 9B model scoring 92.4% on MADQA. October 7, 2026 Perplexity has released pplx-embed-v2-late, a pair of ColBERT-style multimodal embedding models. They come in 2 sizes: 0.6B for fast, cheap queries and 9B for maximum quality. Both models retrieve text, images and rendered PDF pages, and they share one embedding space. Is it deployable? Yes, if you host it yourself. Both models are on Hugging Face under the MIT license. A hosted Perplexity API endpoint is planned but not live. Tl;dr. The best. * The 0.6B model uses about 340M active parameters for images and stays close to 8B rivals. * A 9B index can be searched with 0.6B queries, recovering about half the 9B quality gap on text at 0.6B query cost. * Its 128-dim token vectors are 16x to 32x narrower than rivals at 2,048 to 4,096 dims. * MIT license, with commercial use allowed. The worst. * It stores 1 vector per token, so index size grows with document length. * It is not #1 on ViDoRe v3 image retrieval; Tencent's EVIE scores higher. * A single input cannot mix text and images. * All scores are self-reported, and the technical report is not out yet. Model size and what it runs on. Perplexity designed the 0.6B model as a lightweight query encoder that can also run on edge devices. Both model cards show CUDA GPU usage. They need sentence-transformers >= 6.0.0 and transformers >= 5.4.0. The memory figures are its estimate at 2 bytes per parameter, for weights only. The published checkpoints are stored in F32, which doubles the download size. How well it performs: best and worst scores. All numbers below are from Perplexity's announcement: The strongest result: 92.4% on MADQA, set by the 9B model. The biggest margin is on BrowseComp+, at 8.7pp over the best dense model. The weakest result: 61.2% on ViDoRe v3 Markdown, from the 0.6B model. It is still the 2nd-best score on that benchmark. Image search is the real gap: Gemini Embedding 2 beats the 9B model on MIRACL-Vision and by 2pp on PPLX-Q2I. Mixing sizes: a 9B index queried by the 0.6B model scored 63.5% on ViDoRe v3 image retrieval. That beats 62.3% with 0.6B on both sides, at the same query cost. How it works. Dense models compress a document into 1 vector. pplx-embed-v2-late instead keeps a 128-dim vector for every token. It scores with MaxSim: each query token finds its best document token, and those maxima are summed. Pages are encoded as images, so no OCR step is needed. Perplexity distilled both models from an 18B teacher using LEAF-style token-level training. That training is what creates the shared space. Best use cases. * The best fit is visual document search over PDFs, slides and scanned reports. * Low-latency search: index in the cloud with 9B, then query on-device with 0.6B. * Agentic RAG over large PDF or web collections. Interactive explainer. How it compares. Key takeaways. * The 0.6B model fits edge devices; the 9B model is built for index-time quality. * Best score: 92.4% on MADQA. Weakest: 61.2% on ViDoRe v3 Markdown. * 0.6B queries over a 9B index beat 0.6B on both sides. * Storage growth and self-reported scores are the main caveats.
HP Inc. (HPQ) Stock: Climbs as HP Launches New AI PCs With NVIDIA. Tldr. Table of Contents * HPQ stock climbs 1.61% to $32.24 as HP expands its NVIDIA-powered AI PC lineup. * HP unveils new NVIDIA-powered Windows systems for consumers and enterprise users. * HP demonstrates its ZGX Fury workstation running Windows on NVIDIA GB300 hardware. * HP sets October 16 launch for OmniBook Ultra 16 and OmniBook X 14 AI laptops. * HP adds Perplexity features and a free Pro trial to eligible consumer devices. HP Inc. (HPQ) shares gained 1.61% to close at $32.24, supported by developments in the company's artificial intelligence portfolio. The stock advanced another 0.19% to $32.30 in after-hours trading. Meanwhile, HP expanded its NVIDIA-powered computer lineup and announced new Windows systems for consumers and businesses. HP expands NVIDIA AI portfolio with new Windows systems. HP introduced new computing products that support artificial intelligence tasks directly on personal computers and workstations. The expanded portfolio combines NVIDIA processors with Windows software to serve consumers, developers, and enterprise customers. Additionally, HP outlined plans to increase local processing capabilities across its hardware lineup. The announcement builds on HP's existing efforts to expand beyond traditional personal computers and printing products. HP now targets customers who need stronger computing performance for software development, content creation, and workplace applications. Consequently, the company has expanded its hardware offerings to support different computing requirements. HP also introduced new software features through its partnership with Perplexity. The integration allows users to search for information, complete tasks, and create content through supported consumer devices. Furthermore, HP plans to extend these services across eligible computers beginning in November 2026. HP demonstrates NVIDIA GB300 workstation running Windows. HP demonstrated Windows running on its ZGX Fury workstation powered by NVIDIA's GB300 Grace Blackwell Ultra processor. Microsoft CEO Satya Nadella and NVIDIA CEO Jensen Huang presented the system during a Microsoft event. The demonstration marked HP's first public presentation of the workstation operating within the Windows environment. The ZGX Fury targets developers, researchers, data scientists, and businesses that require substantial computing capacity. Its design supports demanding applications, larger models, and multiple simultaneous requests within local computing environments. As a result, organizations can process sensitive workloads internally while maintaining greater control over their infrastructure. HP previously outlined its GB300 workstation plans during Computex earlier in 2026. The latest demonstration advances those plans by showing how the hardware operates with Microsoft's software platform. However, HP has not announced public pricing or a general release date for the workstation. HP sets October 16 release for new OmniBook AI pcs. HP plans to release two NVIDIA-powered OmniBook laptops on October 16, 2026, targeting users who require advanced computing performance. The OmniBook Ultra 16 will start at $3,199.99, while the OmniBook X 14 will cost $2,999.99. Both devices use NVIDIA RTX Spark technology to support local processing, graphics performance, and creative applications. The company will sell the OmniBook Ultra 16 through its website and Best Buy stores. Meanwhile, customers can purchase the OmniBook X 14 through HP, Costco, Micro Center, Walmart, and other retailers. These launches follow HP's September presentation of its RTX Spark-powered computers at the IFA technology exhibition. The new laptops combine processing power, graphics capabilities, and shared memory to handle demanding tasks without constant cloud access. HP will also offer eligible customers a free one-month Perplexity Pro trial beginning in November. The expansion strengthens HP's position in premium computing as manufacturers introduce more powerful Windows devices for artificial intelligence applications. Limited Time Offer Get 3 free stock ebooks. Discover top-performing stocks in AI, Crypto, and Technology with expert analysis. * Top 10 AI Stocks - Leading AI companies * Top 10 Crypto Stocks - Blockchain leaders * Top 10 Tech Stocks - Tech giants
Perplexity open sources its ColBERT style search models under MIT license. On research.perplexity.ai, the company published pplx-embed-v1, pplx-embed-context-v1, and a late-interaction variant built on the ColBERT approach, all under an MIT license on Hugging Face. That license matters more than the technical specs. Perplexity released open retrieval models tied to the same embedding stack used in its answer engine, free, to developers who want to build search products on their own infrastructure. On research.perplexity.ai, the company published pplx-embed-v1, pplx-embed-context-v1, and a late-interaction variant built on the ColBERT approach, all under an MIT license on Hugging Face. That license matters more than the technical specs. It means a startup can download the weights, inspect them, modify them, and ship a commercial product without paying Perplexity a cent or asking permission. The headline numbers back up the release. According to Perplexity's own research post, the flagship 4B-parameter model scores 69.66 on the MTEB multilingual retrieval benchmark, edging out Alibaba's Qwen3-Embedding-4B at 69.60 and beating Google's gemini-embedding-001, which came in at 67.71. The-decoder.com, which covered the release in detail, noted the models also beat out embeddings from Anthropic and Voyage on MTEB and the ConTEB benchmark, while using a fraction of the memory those rivals need thanks to INT8 and binary quantization. The more interesting model in the release is the late-interaction one, pplx-embed-v1-late-0.6b. Most embedding systems squash an entire document into a single vector, which is fast but throws away detail. A ColBERT-style model instead keeps a 128-dimensional vector for every token and scores a match using MaxSim: for each word in your query, it finds the best-matching word in the document, then adds those scores up. That preserves the kind of fine-grained precision that matters when a query is long, technical, or only partially answered by a given passage. Perplexity says the model beats ColBERT-zero on the BEIR benchmark, 56.61 to its rival's score, and outperforms Jina's jina-colbert-v2 on the multilingual MIRACL benchmark, 66.62 to its rival's number. Embeddings are the unglamorous plumbing behind every retrieval-augmented generation system, every enterprise search box, and every AI product that needs to find the right paragraph before it generates an answer. Many of the highest-profile commercial options still sit behind paid APIs from OpenAI, Cohere, Google, Anthropic and Voyage, each charging per token and keeping its weights closed. Developers already had strong open alternatives such as Qwen3-Embedding, but Perplexity is adding MIT-licensed weights from a company whose core product depends on search quality. IREN signed $2.8 billion in new multi-year AI cloud contracts with Perplexity, Figure AI, Together AI and others, pushing its year-end 2026 AI Cloud revenue target from $3.7 billion to more than $4 billion. The deals build on IREN's earlier $9.7 billion Microsoft and $3.4 billion Nvidia agreements as the former Bitcoin miner completes its pivot to... - AI cloud infrastructure revenue deals - IREN AI cloud computing contracts Perplexity just sharpened that tradeoff. A developer can now run a model that matches Qwen3 and beats Gemini on Perplexity's reported multilingual retrieval benchmark, inspect exactly how it works, fine-tune it on private data, and run it on their own hardware without a per-token API fee. For an AI startup burning through seed funding, that's not a nice-to-do. It's the difference between an embedding bill that scales with usage and infrastructure costs it can control directly. It's also a pointed move against the companies Perplexity competes with directly for search traffic. Google and OpenAI both guard their best retrieval infrastructure closely, because the same tech that ranks your enterprise documents also ranks the web pages their own answer engines pull from. By publishing its models on Hugging Face instead of folding the gains quietly into its own product, Perplexity is betting that giving developers free, inspectable infrastructure builds more goodwill, and more eventual lock-in to its ecosystem, than keeping it proprietary ever would. Open-sourcing core retrieval tech has become something of a pattern this year, not a one-off gesture. Mistral, DeepSeek and Alibaba have all used free, high-quality model releases to win developer mindshare away from closed labs, and Perplexity's embedding drop reads as the same playbook applied to search infrastructure specifically. Whether it actually dents demand for OpenAI's and Cohere's embedding APIs will depend on how fast the ecosystem around these weights grows. But the models are already live, the license is already permissive, and the benchmark comparisons are published for developers to test against their own workloads. That's a real shot, not a marketing claim. Bloomberg found Stable Diffusion, Flux, and Chinese models like Qwen and Hunyuan turning up in active child abuse cases, since offline fine-tuning sidesteps hash-matching tools from Hive and Thorn entirely. - open source AI models child abuse images - how predators use offline AI image generation Join the discussion. No replies yet. Start the discussion. Janet Harrison has over 16 years experience in the financial services industry giving her a vast understanding of how news affects the financial markets, and an early adopter of blockchain technology and digital currencies. Janet is an active holder and trader spending the majority of her time analyzing blockchain projects, reports and watching new and upcoming projects and other initiatives in the industry. She has a Masters Degree in Economics with previous roles counting Investment Banking.
Rings AI and Perplexity have partnered to integrate team relationship intelligence into Perplexity Computer. The collaboration enables joint customers to access their organisation's relationship and activity history within Perplexity Computer workflows. Perplexity Computer handles multi-step tasks across various tools and applications. Rings adds contextual information held across teams, including existing relationships, past meetings and emails, notes, opportunities, and introduction paths through colleagues and shared contacts. Rings captures activity automatically from email and calendar, creating a shared relationship graph. Its PathPower technology measures relationship strength across this graph. The connector supports various use cases, including identifying team members' connections with target companies, prioritising prospects based on relationship strength, and managing existing client relationships. The Rings connector is now available in the Perplexity Computer connector catalogue for Pro, Max, Enterprise Pro, and Enterprise Max subscribers.