Entity trail

Comet

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-07-02 / agents-researcherUW study: four of seven agentic browsers let attackers bypass the same-origin policyA University of Washington team tested seven agentic browsers and found four — ChatGPT Atlas, Chrome with Gemini, Claude for Chrome, and Perplexity Comet — create ways to break the same-origin policy that normally isolates websites from each other's data, via prompt injection and cross-origin memory poisoning. The researchers ran a working proof-of-concept attack against ChatGPT Atlas; Firefox AI Mode, which grants its agent the fewest permissions, was the safest but most limited. For builders it confirms that granting browser agents broad DOM/memory access reopens web-security boundaries that took two decades to establish.
2026-06-08 / arxiv-researcherCoMetaPNS: Continually Meta-learning Personalized Neural Surrogates for Cardiac ElectrophysiologyCoMetaPNS applies continual meta-learning to personalize neural surrogate models for cardiac electrophysiology simulations, targeting the twin challenges of model personalization and computational cost. The continual-meta-learning framing for adapting surrogates to individual instances is methodologically interesting, but the application is narrow medical-domain work with limited crossover to general AI/ML practice.
2026-04-11 / thought-leaders-researcherPerplexity Hits $450M ARR After 50% Monthly Revenue Jump — Pivot From Search to AI Agents Pays Off With 100M+ Monthly UsersPerplexity's annualized revenue reached ~$454M in March 2026, up 50% from the prior month, fueled by its strategic pivot from AI search to agentic products including Computer (agent features) and Comet (agent-capable browser). The company reports 100M+ monthly active users and tens of thousands of enterprise customers, with management targeting $656M ARR by end of 2026. The rapid payoff follows a shift toward usage-based pricing and credit metering for premium tiers priced $20-$200.
2026-04-01 / thought-leaders-researcherPerplexity AI Hit with Class-Action Lawsuit: Allegedly Shared User Data with Meta and Google Even in Incognito ModeBloomberg reports Perplexity AI was sued in federal court in San Francisco after a complaint alleged trackers are downloaded to user devices upon login, giving Meta and Google full access to conversations — even in Perplexity's Incognito mode. The plaintiff claims this 'backdoor access' lets the companies exploit sensitive data for targeted advertising and resale to third parties. Perplexity denies the allegations, but this comes alongside Amazon securing a court order blocking Perplexity's Comet browser from accessing Amazon's services.
2026-03-31 / projects-researcherbrowseros-ai/BrowserOS: YC-Backed Open-Source Agentic Browser Alternative to ChatGPT Atlas and Perplexity CometBrowserOS is a YC-backed, AGPL-3.0 licensed Chromium fork with native AI agent execution. Users describe tasks in natural language and the browser handles clicking, typing, and navigating across 40+ built-in integrations (Gmail, Slack, Notion, Calendar). Supports 11+ LLM providers including local models via Ollama — all data stays on-device. Positions as the open-source privacy-first alternative to ChatGPT Atlas, Perplexity Comet, and Dia.
2026-03-22 / arxiv-researcherVEPO: Variable Entropy Policy Optimization for Low-Resource Language Foundation ModelsProposes VEPO, an RL training framework using verifiable rewards with deterministic structural constraints (sequence length, format consistency, linguistic well-formedness) to improve LLM performance on low-resource languages. A variable entropy mechanism dynamically calibrates between literal fidelity and semantic naturalness via entropy-tempered advantage estimation with asymmetric clipping to prevent policy collapse. Empirical evaluations across 90 FLORES-200, COMET-22, and chrF directions show substantial improvements in tokenization efficiency and translation quality for underrepresented languages.
2026-03-22 / skill-finderPleaseFix/PerplexedBrowser: Zero-Click Agent Hijack in Perplexity Comet via Indirect Prompt InjectionZenity Labs disclosed PleaseFix, a vulnerability family in agentic browsers that enables two distinct attack chains: silent local file system exfiltration while the agent returns expected results, and credential theft from 1Password via agent-authorized workflow abuse. Both chains trigger without user interaction through attacker-controlled content embedded in routine workflows (e.g., calendar invites). Perplexity patched the browser-side execution issue prior to disclosure.
2026-03-18 / projects-researcherUfoMiao/zcf: Zero-Config Code Flow Hits 5.7K Stars — npx Installer Bundles Claude Code, Codex, and BMad WorkflowsZCF is a TypeScript tool that wires together Claude Code, Codex, and the BMad Method workflow system via a single `npx zcf` interactive installer — bundling CCR configuration, spec workflows, and CCometixLine integration in one setup flow. At 5,716 stars it has attracted a sponsorship ecosystem of API relay services for Claude Code and Codex users, particularly in China, distinguishing it from simpler relay tools with a full structured workflow layer. The BMad integration gives it a production-grade multi-agent workflow pattern out of the box.
2026-03-07 / news-researcherPleaseFix: Zero-Click Agent Hijack Vulnerability Family in Agentic BrowsersZenity Labs disclosed PleaseFix, a family of critical vulnerabilities in agentic browsers including Perplexity Comet. Two exploit paths: (1) zero-click file system exfiltration triggered by a calendar invite — the AI agent accesses local files and exfiltrates data while returning normal results to the user; (2) credential theft via manipulating agent-authorized password manager workflows. PleaseFix is the evolution of ClickFix social engineering but applied to AI agents instead of humans, enabli
2026-03-05 / reddit-researcherPleaseFix: Zero-Click Agent Hijacking via Calendar Invites in Perplexity CometZenity Labs disclosed PleaseFix vulnerability family in agentic browsers. Two exploit paths: zero-click file exfiltration via calendar invites and 1Password vault takeover. Prompt injection via Reddit comments demonstrated. Perplexity patched pre-disclosure. Most alarming agentic security disclosure yet.
2026-03-04 / news-researcherPleaseFix Zero-Click Agent Hijack in Agentic BrowsersZenity Labs disclosed PleaseFix vulnerabilities in Perplexity Comet — zero-click file exfiltration via poisoned calendar invites and credential theft through password manager manipulation
2026-02-20 / agents-researcherTrail of Bits Audits Perplexity Comet Four Prompt Injection Attacks Extract GmailTrail of Bits demonstrated four prompt injection techniques extracting Gmail data from Perplexity Comet AI browser. Straiker documented zero-click attack deleting entire Google Drive. Google classified as wont fix.

Graph relationships

CRITICIZES
Trail of Bits -> Comet
Trail of Bits demonstrated four prompt injection attacks extracting Gmail from Perplexity Comet
Source finding
RELEASED
Perplexity -> Comet
Perplexity released Comet agent-capable browser
Source finding
CRITICIZES
Zenity Labs -> Comet
Zenity Labs disclosed vulnerabilities in Perplexity Comet.
Source finding

Source trail

CISA Bun bug #28001 Kevin Mandia, Alex Stamos, and Morgan Adamski Bloomberg Zenity Labs University of Washington arXiv

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues