AI Built a $250,000 Chrome Exploit in 6 Days: What GPT-5.6 Sol Means for Chrome Users

Artificial intelligence helped security researchers build a Google Chrome exploit chain worth a $250,000 bug bounty in just six days, offering one of the clearest examples yet of how rapidly AI is changing cybersecurity.

Security company XBOW disclosed the research on October 13, 2026.

The company says one of its researchers worked with AI agents using OpenAI’s GPT-5.6 Sol to combine three previously unknown Chrome vulnerabilities into a full exploit chain.

Before Google patched the bugs, XBOW estimated that the attack path could have affected roughly:

1 billion Chrome users on Windows.

The resulting chain could escape Chrome’s security sandboxes and run code on the underlying Windows computer simply after a malicious webpage was loaded.

Google verified the findings, fixed the vulnerabilities and awarded XBOW:

$250,000.

That is Google’s top reward for a qualifying Chrome full-chain exploit under its current vulnerability-reward rules.

But there is also an extremely important fact for ordinary Chrome users:

XBOW says the vulnerabilities were never used against Chrome users in the wild.

All three bugs involved in the chain have been fixed.

So the story is not that GPT-5.6 Sol secretly hacked one billion computers.

The real story is arguably more important:

AI dramatically accelerated work that once required elite security teams months to perform.

Here is what happened, how GPT-5.6 Sol was used, why Google paid $250,000 and what Chrome users should do now.

What Happened With the GPT-5.6 Sol Chrome Exploit?

XBOW set itself a difficult security-research challenge:

build a complete exploit chain against Google Chrome.

Chrome is intentionally difficult to compromise because it uses multiple security boundaries.

Finding one vulnerability is therefore usually not enough.

An attacker would need several bugs that work together.

XBOW’s final awarded chain combined vulnerabilities in three separate Chrome components:

Stage	Chrome component	Result
1	V8 JavaScript engine	Gained control inside an inner sandbox
2	Animated WebP processing	Helped cross another security boundary
3	Crashpad	Enabled code execution outside Chrome’s sandbox

The result demonstrated a path from:

visiting a webpage

to:

running code outside Chrome’s security sandbox on Windows.

XBOW says the user would not have needed to intentionally download a malicious program or click an additional exploit button.

That is why the chain qualified for such a large reward.

Did AI Really Build the Chrome Exploit in Six Days?

Yes—but the headline needs context.

The awarded exploit chain was built during a six-day sprint from:

August 2 through August 7, 2026

during the Black Hat and DEF CON security conferences in Las Vegas.

However, that does not mean XBOW started the entire project from zero on August 2.

The company had been working on the broader Chrome challenge earlier in the year.

XBOW says its first full-chain attempt took roughly:

two months.

One of the bugs in that earlier chain turned out to be something Google already knew about.

That meant the researchers needed another approach.

For the final awarded chain, XBOW had designed the components so alternatives could be substituted if one vulnerability was disqualified.

One researcher, Brendan Dolan-Gavitt, then worked with AI agents to assemble the replacement chain in:

six days.

That six-day chain is the one that ultimately earned the $250,000 reward.

What Did GPT-5.6 Sol Actually Do?

This was not simply a human researcher asking a chatbot:

“Please hack Chrome.”

XBOW describes a much more sophisticated workflow.

AI agents using GPT-5.6 Sol helped:

• search Chrome’s source code;
• investigate possible vulnerabilities;
• test ideas;
• analyze failures;
• develop proof-of-concept inputs;
• draft early exploit components;
• continue research across long sessions.

Human researchers still made crucial decisions.

They:

• chose the overall strategy;
• decided which areas of Chrome were worth investigating;
• reviewed the agents’ findings;
• rejected weak leads;
• redirected the AI;
• selected which vulnerabilities to combine;
• validated the final chain.

So the most accurate description is:

human-directed, AI-accelerated vulnerability research.

That distinction matters.

GPT-5.6 Sol did not independently decide to attack Google Chrome.

Researchers intentionally used it in an authorized security-research program and reported the vulnerabilities to Google.

How Much Did the AI Research Cost?

XBOW estimates that the compute used during the six-day awarded-chain sprint cost approximately:

$40,000.

Google’s reward was:

$250,000.

That means the bounty itself was more than six times the reported AI compute cost.

However, the $40,000 number should not be interpreted as the total cost of the research project.

It does not account for:

• researcher salaries;
• prior research;
• infrastructure;
• engineering;
• earlier unsuccessful work;
• company operating costs.

The important comparison is instead how quickly AI helped perform extremely difficult security research.

XBOW says work like this historically could take an elite team:

many months.

The awarded chain was assembled in:

six days.

Why Did Google Pay $250,000?

Google’s Chrome Vulnerability Reward Program pays researchers who responsibly disclose serious security weaknesses.

The largest rewards are reserved for attack chains capable of crossing Chrome’s major security boundaries.

A normal browser vulnerability might compromise only the webpage process.

Chrome’s sandboxes are intended to prevent that compromise from reaching the broader computer.

A full chain defeats multiple layers.

XBOW demonstrated an attack path that ultimately reached:

code execution outside the browser sandbox.

That qualified for Google’s highest full-chain bounty:

$250,000.

XBOW says it was Google’s first $250,000 Chrome full-chain reward under the 2026 rules.

What Is a Chrome Full-Chain Exploit?

A full-chain exploit combines several vulnerabilities.

Think of Chrome security as a building with several locked doors.

Finding one vulnerability might open:

Door 1.

But the attacker is still trapped behind:

Door 2

and:

Door 3.

A full-chain exploit finds a way through each boundary.

In XBOW’s demonstration, the chain moved through Chrome components including:

V8 → image processing → Crashpad

until code could execute outside the normal browser sandbox.

That is substantially more serious than a single isolated browser crash.

Which Chrome Vulnerabilities Were Involved?

Two publicly identified vulnerabilities associated with the chain include:

CVE-2026-85045

and:

CVE-2026-78952.

Google’s release notes identify CVE-2026-85045 as a high-severity race condition in:

V8

reported by XBOW researcher Brendan Dolan-Gavitt.

Google identifies CVE-2026-78952 as a high-severity:

out-of-bounds write in Crashpad.

The chain also involved an animated-WebP processing issue.

Because public descriptions across the initial disclosures have differed regarding the identifier associated with that intermediate stage, it is safer to rely on Google’s own release notes for CVE mappings rather than repeating an uncertain identifier.

The important user-facing point is:

Google fixed the entire chain.

Was the Exploit Used Against Chrome Users?

XBOW says:

No.

The company says the vulnerabilities were never used against Chrome users.

They were discovered through authorized defensive research and disclosed to Google.

Google then reproduced the findings and released patches.

There is currently no evidence from the disclosed research that criminals used this specific three-bug XBOW chain against ordinary users before it was fixed.

That distinction is extremely important.

A vulnerability being capable of affecting a billion users does not mean one billion people were hacked.

It means the vulnerable software existed across a very large potential user population before patches were installed.

Was One Billion Chrome Users Really at Risk?

XBOW estimates that approximately:

1 billion Chrome users on Windows

were potentially in the path of the vulnerability class before patches.

That number is an estimate from XBOW rather than a count of compromised devices.

Again:

potentially exposed ≠ successfully hacked.

The story is significant because Chrome is one of the world’s most widely used browsers and because the exploit chain targeted a highly hardened piece of software.

Do Chrome Users Need to Panic?

No.

The vulnerabilities have already been fixed.

If your Chrome browser is current, you are well beyond the original affected releases.

The two publicly identified vulnerabilities were fixed during the Chrome 152 release cycle in August and September 2026.

Chrome has since advanced several major versions.

As of October 20, Google had promoted:

Chrome 156

to the stable desktop channel.

Some users were already beginning to receive early Chrome 157 builds by October 22.

Therefore the practical advice is simple:

update Chrome and restart it.

You do not need to install a special XBOW patch.

How to Check Whether Chrome Is Updated

On desktop Chrome:

1. Open Chrome.
2. Open the three-dot menu.
3. Choose Help.
4. Select About Google Chrome.
5. Allow Chrome to check for updates.
6. Relaunch Chrome if prompted.

The relaunch step matters.

Chrome can download an update in the background, but the new version may not fully take effect until the browser restarts.

If you routinely leave Chrome open for weeks with dozens of tabs, restart it periodically so security updates can finish installing.

Does Antivirus Protect Against This Kind of Chrome Exploit?

Security software can provide additional layers of protection, but antivirus is not a substitute for updating the vulnerable browser.

A browser vulnerability exists inside the software itself.

The best first defense is:

install the vendor’s security update.

Additional protection can include:

• current operating-system patches;
• endpoint security;
• safe browsing protections;
• limited administrator privileges;
• secure browser settings.

But browser updates remain essential.

Did GPT-5.6 Sol Autonomously Hack Chrome?

No.

This is an important misconception to avoid.

OpenAI’s own published evaluations of GPT-5.6 Sol found that the model could identify vulnerabilities and exploitation building blocks in browsers such as Chromium and Firefox.

But under OpenAI’s controlled evaluations, GPT-5.6 Sol did not autonomously produce a complete functional full-chain exploit.

The XBOW result demonstrates something slightly different.

The model became substantially more capable when paired with:

• experienced security researchers;
• autonomous-agent infrastructure;
• build systems;
• debugging tools;
• verification;
• human strategic judgment.

That combination matters.

The strongest lesson is not:

“AI can independently hack Chrome.”

It is:

“AI plus skilled researchers can dramatically accelerate advanced vulnerability research.”

Why Is This Important for Cybersecurity?

Security teams traditionally operate under an important assumption:

finding and exploiting the hardest vulnerabilities requires enormous amounts of specialist time.

AI is beginning to change that.

Frontier models can increasingly assist with:

• source-code review;
• crash analysis;
• vulnerability discovery;
• reproducing failures;
• debugging;
• reasoning across large codebases;
• patch analysis.

That reduces the time required to investigate potential vulnerabilities.

The same shift appeared in the recent PHP2Shell research, where AI played an important role in finding and validating serious vulnerabilities affecting PHP-powered websites.

Read our PHP2Shell coverage for the broader website-security implications.

AI Is Lowering the Cost of Vulnerability Research

The XBOW result follows a broader pattern.

Elite Era Trends recently covered a separate campaign in which autonomous AI agents were used to target more than:

1,200 websites

and gain access to sites associated with more than:

100 companies.

That campaign showed how agents can reduce the cost of:

large-scale attack automation.

The Chrome research shows the other side:

AI can also dramatically accelerate:

high-end defensive vulnerability discovery.

These trends are related but not identical.

One involves malicious actors using automation.

The other involves authorized researchers discovering vulnerabilities and reporting them so they can be fixed.

The technology can benefit both:

attackers and defenders.

What Does This Mean for Agentic AI?

XBOW’s work is also a good example of:

agentic AI.

An ordinary chatbot typically follows:

prompt → response.

An AI research agent can instead work through something closer to:

goal → investigate → test → observe → adjust → test again.

That makes the system much more useful for long cybersecurity tasks.

Our guide to What Is Agentic AI? explains how this kind of multi-step autonomy differs from ordinary generative AI.

Is GPT-5.6 Sol Still OpenAI’s Main Model?

No.

GPT-5.6 Sol remains important historically because it powered the XBOW research.

But OpenAI has since moved its mainstream ChatGPT model family forward.

Elite Era Trends covers the newer GPT-6 family, including GPT-6 Astra, which OpenAI positions for difficult reasoning, coding, professional work and computer use.

The XBOW story is therefore best understood as evidence of how rapidly model cyber capabilities were already advancing before the latest generation arrived.

Does This Mean Newer AI Will Be Even Better at Hacking?

Not necessarily in a simple linear way.

More capable AI models can become better at legitimate security tasks such as:

• finding vulnerabilities;
• reviewing secure code;
• writing patches;
• investigating malware;
• analyzing incidents.

Those capabilities can also create misuse risks.

AI companies therefore increasingly use:

• account verification;
• cyber-specific safeguards;
• usage monitoring;
• model-level restrictions;
• access controls.

OpenAI classifies GPT-5.6 Sol as having high cybersecurity capability but says it did not cross the company’s highest Cyber Critical threshold in its controlled evaluations.

Capability and safety therefore have to develop together.

Should Powerful Cyber AI Models Be Restricted?

This is now an active debate.

One argument says that more capable models could make sophisticated attacks cheaper and faster.

Another says defenders need access to the same capabilities because attackers will eventually obtain comparable tools anyway.

XBOW’s position is largely defensive:

use AI to discover vulnerabilities before criminals do.

That approach can create a race:

AI attacker vs. AI defender.

The goal for software companies is to make sure defensive research happens first.

What Chrome Users Should Do Now

For ordinary users, the response is straightforward.

1. Update Chrome

Go to:

Chrome → Help → About Google Chrome

and install any available update.

2. Relaunch the Browser

Do not indefinitely postpone the restart.

3. Keep Windows Updated

The demonstrated exploit chain targeted Chrome on Windows.

Keep the operating system current as well.

4. Be Careful With Extensions

Browser extensions create another security surface.

Our BragJack coverage explains how researchers demonstrated attacks in which malicious extensions could interfere with AI browser assistants.

Install only extensions you need and trust.

5. Keep Safe Browsing Enabled

Chrome’s built-in protections can help warn about known malicious websites and downloads.

6. Don’t Assume “I Didn’t Click Anything” Means You’re Always Safe

The XBOW research is a reminder that some browser vulnerabilities can potentially trigger through webpage processing itself.

That is why browser patching matters even for cautious users.

AI Browser Security Is Becoming a Bigger Topic

Chrome security is no longer only about traditional memory vulnerabilities.

Browsers are also gaining:

• AI assistants;
• computer-use capabilities;
• autonomous agents;
• connected apps;
• browser-control tools.

That creates new security problems.

Our guide to AI Browser Prompt Injection explains how malicious webpage content can attempt to manipulate browser agents.

Our BragJack article covers a different attack class involving malicious browser extensions and trusted AI-control interfaces.

Together, these developments show why browser security now has two layers:

traditional software security

and:

AI-agent security.

GPT-5.6 Sol Chrome Exploit: Key Facts

Fact	Details
Research organization	XBOW
AI model	GPT-5.6 Sol
Award	$250,000
Awarded-chain build time	6 days
Estimated compute cost	$40,000
Vulnerabilities chained	3
Target	Google Chrome on Windows
Estimated potential users affected	~1 billion
Used against users in the wild?	XBOW says no
Google fixed the bugs?	Yes
User action now	Update and relaunch Chrome

Frequently Asked Questions

Did GPT-5.6 Sol hack Google Chrome?

GPT-5.6 Sol helped XBOW researchers discover, test and develop parts of a Chrome exploit chain. Human researchers directed the work, reviewed results and made key strategic decisions.

How much did Google pay XBOW?

Google awarded XBOW $250,000 for the qualifying Chrome full-chain exploit.

How long did it take?

The final awarded chain was assembled during a six-day sprint in August 2026. XBOW had performed earlier research and built an earlier chain before the six-day effort.

How much did the AI cost?

XBOW estimates approximately $40,000 in compute during the six-day awarded-chain effort.

How many Chrome vulnerabilities were used?

The exploit combined:

three vulnerabilities.

What were the Chrome vulnerabilities?

The chain involved Chrome’s V8 JavaScript engine, animated-WebP processing and Crashpad. Publicly identified vulnerabilities associated with the chain include CVE-2026-85045 and CVE-2026-78952.

Did hackers exploit these bugs against real Chrome users?

XBOW says the vulnerabilities in its chain were not used against Chrome users.

Were one billion users hacked?

No. XBOW estimates roughly one billion Windows Chrome users were potentially exposed before the bugs were fixed. That is not the same as one billion compromised devices.

Are the Chrome vulnerabilities fixed?

Yes.

Google confirmed and patched the vulnerabilities.

Should I update Chrome?

Yes. Keeping Chrome current is one of the most important steps you can take to protect against known browser vulnerabilities.

How do I update Chrome?

Go to:

Chrome → Help → About Google Chrome

and relaunch the browser after the update installs.

Can AI now autonomously create full Chrome exploits?

The XBOW result does not demonstrate a completely autonomous model doing so from beginning to end. Human security researchers played an essential role in strategy, review and integration.

Why is this story important?

It shows that AI agents can dramatically accelerate advanced vulnerability research when paired with skilled human researchers and security tooling.

Bottom Line

The $250,000 XBOW Chrome exploit is one of the clearest signs yet that AI is changing cybersecurity.

A researcher working with GPT-5.6 Sol-powered agents assembled the final three-bug Chrome exploit chain in:

six days.

XBOW estimates the AI compute cost at:

about $40,000.

Google paid:

$250,000

after confirming the vulnerabilities.

Before they were patched, XBOW estimated the attack path could have affected approximately:

one billion Windows Chrome users.

But ordinary users should understand the equally important part of the story:

the vulnerabilities were responsibly disclosed, Google fixed them and XBOW says they were never used against Chrome users.

There is no reason to panic.

There is a reason to update.

The larger cybersecurity lesson is more significant.

AI is lowering the amount of time required to perform work that once demanded months from elite specialists.

That can help defenders find dangerous vulnerabilities sooner.

It can also reduce the time software companies have to respond before similar capabilities reach malicious actors.

Cybersecurity is therefore entering a new race:

AI finding vulnerabilities

versus:

AI helping fix them.

For Chrome users, the immediate action remains simple:

keep Chrome updated and restart it when an update is ready.