Reported by 2 outlets
1. Basic Information Original Title: BragJack [Technical Overview]: How We Hijacked Top 5... # security # threatintel # software # coding #
Why this score
- Breadth
- 1.00 / 2.0 · 2 sources
- Authority
- 1.00 / 2.0 · best outlet weight
- Severity
- 0.50 / 5.0 · CVSS + content signals
- Velocity
- 0.00 / 2.0 · updates in 48h
- Pickup speed
- 0.60 / 0.6 · time to second source
- Recency
- × 0.00 temporal multiplier
- Community
- + 0.00 / 1.5 · reddit signal
- Score
- 0.0 / 10 = base × recency + community
Analysis
BragJack is a prompt-forcing technique that uses malicious browser extensions to hijack the AI agents now embedded in the five major browsers. The dev.to technical overview by the researcher behind the name walks through how an extension, once installed, can inject instructions into the agent's context so the assistant follows attacker commands instead of user intent, and cybersecuritynews.com covered the same work as a new prompt-forcing class against agentic browsing. The practical risk is that an agent with authenticated sessions, browsing history, and tool access becomes an extension-controlled proxy for data theft or unauthorized actions.
This story cluster is broader than BragJack itself: it aggregates a rolling series of dev.to reposts by the same author alongside the BragJack coverage. Those reposts include KREMLIN malware forging Chromium integrity checks to steal banking sessions, the Iranian Chosen Brick surveillance tool using Telegram C2, the North Korean WaterPlum campaign infecting 30,000 devices through fake interview tasks, and Rapuncel, which pairs fake GitHub repositories with a signed kernel driver to disable EDR. The most recent event extends the AI theme, describing autonomous AI agents breaching online retailers in chained attacks to steal payment card data.
Read it as a signal that agentic AI is becoming a standing attack surface with its own exploit classes. Watch for follow-up research on extension-to-agent injection primitives, and for browser vendors adding isolation between extensions and built-in assistants. For defenders, the actionable checks are extension inventory and provenance in the enterprise, plus limits on what agent sessions may reach: payment flows and authenticated admin surfaces should not be reachable from an assistant that any extension can steer.
Timeline
-
original reporting
1. Basic Information Original Title: BragJack [Technical Overview]: How We Hijacked Top 5... # security # threatintel # software # coding #
dev.to · 26d ago · September 17th, 2026
- Basic Information Original Title: BragJack [Technical Overview]: How We Hijacked Top 5... # security # threatintel # software # coding #
via Mastodon https://dev.to/anoymask/bragjack-prompt-forcing-in-browser-ai-agents-via-browser-extensions-1d67
-
update from dev.to
1. Overview Original Title: Iranian cyber targeting of dissidents, activists and... # security # threatintel # software # coding # developme
dev.to · 26d ago · September 17th, 2026
Overview Original Title: Iranian cyber targeting of dissidents, activists and... # security # threatintel # software # coding # developme
via Mastodon https://dev.to/anoymask/chosen-brick-iranian-windows-surveillance-malware-using-telegram-c2-34ng
-
update from dev.to
1. Basic Information Original Title: The extension you never installed: KREMLIN forges... # security # threatintel # software # coding # dev
dev.to · 26d ago · September 17th, 2026
Basic Information Original Title: The extension you never installed: KREMLIN forges... # security # threatintel # software # coding # dev
via Mastodon https://dev.to/anoymask/kremlin-forging-chromium-integrity-checks-to-steal-banking-sessions-oli
-
update from dev.to
1. Basic Information Original Title: North Korean "WaterPlum," commonly referred to as... # security # threatintel # software # coding # dev
dev.to · 25d ago · September 19th, 2026
Basic Information Original Title: North Korean "WaterPlum," commonly referred to as... # security # threatintel # software # coding # dev
via Mastodon https://dev.to/anoymask/waterplum-north-korean-campaign-infects-30000-devices-via-fake-interview-tasks-4528
-
update from dev.to
1. Basic Information Original Title: Threat Intel | One Kit, Forty Companies: How a... # security # threatintel # software # coding # develo
dev.to · 25d ago · September 19th, 2026
Basic Information Original Title: Threat Intel | One Kit, Forty Companies: How a... # security # threatintel # software # coding # develo
via Mastodon https://dev.to/anoymask/rapuncel-fake-github-repositories-disable-edr-with-a-signed-kernel-driver-1nl8
-
update from cybersecuritynews.com
BragJack attack lets malicious extensions hijack AI agents in five major browsers - new prompt-forcing technique
cybersecuritynews.com · 24d ago · September 19th, 2026
BragJack attack lets malicious extensions hijack AI agents in five major browsers - new prompt-forcing technique
via Mastodon https://cybersecuritynews.com/bragjack-ai-agent-hijacking/