How to check whether your ChatGPT, Claude, or Perplexity account has been hacked

AI chat tools have become ordinary work accounts, which makes them a decent target for the usual mess: stolen cookies, reused passwords, phishing, and opportunistic sign-ins from leaked credentials. TechCrunch recently published a practical check for three of the biggest consumer AI products, and it’s worth knowing if your team uses any of them day to day.

The short version: ChatGPT, Claude, and Perplexity all give you some way to inspect or kill active sessions. ChatGPT and Perplexity support MFA in some form. Claude takes a different route and uses email-based login links instead of passwords, which changes the attack surface but doesn’t remove it.

Why this matters for technical teams

For developers and engineers, an AI account is often more sensitive than a random SaaS login. People paste code, architecture notes, API keys, incident details, and draft product plans into these tools without thinking twice. If someone gets into that account, they don’t just get chat history. They may get a clean feed of internal context.

That’s why session management matters as much as password policy here. A lot of account compromises don’t start with a dramatic password change. They start with a live session stolen from a browser profile, a reused credential, or a malicious login that slips through before MFA enforcement catches up.

The useful part of TechCrunch’s guide is that all three platforms expose session controls somewhere in account settings. The exact path is different, and that’s where users usually get lost.

ChatGPT: session review is straightforward, password reset is a little clunky

In ChatGPT, the path is buried a few clicks deep but still usable. Open ChatGPT in a browser, click your username in the lower-left corner, go to Settings, then Security and Login, then Active Sessions.

That view shows where your account is logged in. If a device looks wrong, you can sign out that single session or use Log out all to wipe everything. That’s the right move if you suspect token theft or if you’re not sure whether the bad actor is still active.

One detail matters: if you want to change your password, you first have to log out of the account. OpenAI then sends a six-digit code by email, which you enter on the login page before setting a new password. It’s clunky, but it keeps password recovery tied to email control rather than a separate recovery app or hardware key flow.

The limitation is obvious. Session lists only help if you know what to look for. If the attacker is using your own laptop, your own office IP, or a VPN endpoint that looks normal enough, the signal gets noisy fast. Treat the session list as one clue, not proof either way.

Claude: passwordless login helps, but email becomes the choke point

Claude works differently because Anthropic doesn’t use a password login here. Instead, it sends a login link to your email address. That cuts down on password reuse problems, which is good. It also means the security of your Claude account depends heavily on the security of your inbox.

If someone gets into your email, they can get into Claude. That’s a familiar trade-off, and it’s a good reason to treat email account protection as infrastructure, not convenience.

To check active sessions in Claude, open it in a browser, click your username in the lower-left corner, then Settings, then Account. You’ll see Active sessions. From there, hover over a session, click the three dots, and choose Log out or Terminate. There’s also an option to log out of all devices.

That session control is the important part. Passwordless login doesn’t stop session theft. If a browser session token has already been copied or hijacked, the attacker can keep using it until you revoke it. Email link auth just moves the weak spot around.

For teams, the lesson is simple: “no password” does not mean “no account risk.” If your org uses Claude heavily, email MFA should be mandatory. Mailbox auditing should be too.

Perplexity: a blunt session reset, which is fine if you know to use it

Perplexity is the least informative of the three. It doesn’t show where you’re logged in, which is a miss. If a platform offers no visible session inventory, users lose a useful way to spot suspicious access patterns.

What it does offer is a nuclear option. Go to Perplexity in a browser, click your username in the lower-left corner, choose All settings, then Sign out of all sessions, and confirm.

That’s simple, but it’s less granular than ChatGPT or Claude. You can’t inspect individual sessions first, so you’re forced to sign everyone out if something looks off. That’s acceptable for a consumer product, less ideal for business use. When an AI tool sits inside day-to-day engineering workflows, admin visibility matters. Session listing, IP history, device metadata, and forced revocation are basic hygiene.

Perplexity also supports MFA, which is good. As always, though, MFA only helps if the implementation and attack path around it hold up. If a session cookie is stolen after MFA succeeds, MFA won’t save you retroactively.

What actually gets compromised

People often focus on passwords because they’re visible. In practice, the bigger risk is usually session persistence.

A lot of modern web apps, including AI platforms, rely on bearer tokens stored in browser cookies or local storage. Once an attacker gets a valid session token, they often don’t need the password at all. They just ride the existing session until it expires or gets revoked. That’s why a visible active-session page is more than a cosmetic feature. It’s one of the few practical tools users have to kill a hijacked login without waiting for support.

The other issue is shared devices and browser profiles. If someone signs into ChatGPT or Claude on a work laptop, leaves the browser profile open, and syncs it across machines, you’ve created a wider session footprint than you probably wanted. The same thing happens with home browsers that keep tabs alive forever. Security controls can’t fix sloppy session hygiene.

A few habits that matter more than people admit

If your team uses these tools for real work, treat the accounts like anything else that touches internal systems.

• Use a password manager for the email account tied to AI logins.
• Turn on MFA anywhere it’s offered.
• Review active sessions after travel, laptop loss, or a phishing scare.
• Kill all sessions if you see anything weird.
• Don’t paste secrets into chats unless you’re comfortable treating that service as part of your data perimeter.
• Keep a separate browser profile for work AI tools if you can.

That last one sounds fussy, but it’s practical. Separate profiles reduce session spillover, make audits easier, and keep personal browser junk from muddying the water when something looks suspicious.

The bigger issue is visibility

The three products here show three different maturity levels, and that says a lot.

ChatGPT gives you decent session inspection and a clear revocation path. Claude does the same, but around a passwordless email-login model. Perplexity gives you a blunt sign-out-all control and less visibility into where you’re logged in. None of that is unusual, but it does affect how confidently users can respond to a suspected compromise.

For regular users, that means checking sessions should be part of basic account hygiene, not a panic move after the fact. For engineering teams, it’s a reminder to ask a simple question before adopting another AI tool broadly: can users and admins actually see and revoke active sessions without filing a ticket?

That question matters more than most vendor security slides ever will.