Browser Security

The average Okta customer now runs 101 apps¹ and staff access those apps through the browser, in sessions outside the view of security.

That increased exposure has made businesses vulnerable.

In FY2024-25, ASD responded to 11% more cyber incidents, and the average self-reported cost of a business cybercrime report rose 50% to $80,850, with compromised accounts or credentials a factor in 42% of the incidents ASD rated C3 or higher.²

Credential theft pays threat actors who sell the data on the dark web to large syndicates.

From session data to sold on the dark web

The browser is where attackers steal credentials and active session tokens. The dark web is where they profit. IBM X-Force’s 2025 Threat Intelligence Index describes a thriving dark web trade in stolen credentials.³

Session hijacking kits are also sold to exploit the browser. Attackers buy adversary-in-the-middle phishing kits⁴ to steal valid tokens. Once an attacker holds a valid session token, they inherit all that user's access without triggering any further authentication challenge.

What "stolen browser data" looks like in practice: In April 2026, Vercel disclosed a breach that began at Context.ai, a third party AI tool one of its staff had trialled. A Context.ai employee’s computer was infected with the Lumma infostealer after they downloaded Roblox cheat scripts, which took Google Workspace, Supabase, Datadog and AuthKit credentials. Context.ai says an attacker then reached its AWS environment and likely took OAuth tokens. One belonged to a Vercel employee who had signed up with a corporate Google account and granted “Allow All” access. The attacker used it to take over the account and move into Vercel’s systems. Nobody broke through Vercel’s perimeter. The same weekend, a seller offered the data for about USD $2 million.⁵

Attackers exploit that gap by cloning popular AI assistant extensions: in January 2026, two fake AI sidebar extensions, one carrying Google’s Featured badge, were found stealing ChatGPT and DeepSeek chats from about 900,000 users. Some legitimate extensions have added similar collection as a product feature: Secure Annex found the Similarweb extension, used by over a million people, began capturing AI prompts and responses after a May 2025 update, disclosed in its privacy policy.⁷ Detection means watching extension behaviour over time, not just at install.

Agentic browsers, the kind that take actions on a user's behalf, read untrusted webpage content as if it were a user instruction. A staff member asks the agent to summarise a Reddit thread. Hidden text in a comment tells it to collect the user’s email address and a one-time login code from Gmail and post them back to Reddit, handing the attacker the account. There is no phishing click and no malware, just the user’s own signed-in session doing the attacker’s work.⁸

Blocking creates an internal risk

Many businesses don't consider the additional risk they create by blocking AI tools.

When the IT team locks down AI tools or restricts SaaS access overnight without guidance, staff route around it, moving to personal devices, personal accounts, or personal subscriptions.

When a user moves from a managed device to an unmanaged personal computer control is lost. It's no surprise attackers are engineering attacks that take advantage of this tension.⁹ Among infostealer-infected systems holding corporate logins, 46% were unmanaged devices, outside the reach of corporate endpoint tools.¹⁰