How did GPT-5.6 Sol Ultra find a WordPress zero-day?

Adam Kues adapted OpenAI's math-conjecture prompt into a bug-hunt harness that ran up to four Sol Ultra agents in parallel over about six hours, restricted from web search and git history so the model had to reason from source. It surfaced a bypass in WordPress's batch API that anchored the RCE chain.

How much did the WordPress RCE cost to find?

Searchlight Cyber estimates about $25 of GPT-5.6 Sol Ultra API time, roughly half of a $200 weekly subscription. Kues put in about ten hours of human work on top of that — for a bug class that dark-market exploit brokers advertise at $500,000.

Which WordPress sites are affected by the GPT-5.6 exploit?

Searchlight Cyber says the pre-auth chain works against a default WordPress install and estimates 500 million+ sites are exposed. Patches are being coordinated; site operators should apply the current WordPress security release as soon as it lands and audit for the batch-API and embed vectors described in the write-up.

Does this mean AI can now find zero-days on its own?

Not on its own — but close. Adam Kues framed GPT-5.6 Sol Ultra as a force multiplier: parallel agents chewed through code paths a human would take days to trace, and the human still stitched the chain together, judged exploitability and did coordinated disclosure. It is the strongest public demo yet of AI-assisted vulnerability research.

Searchlight Cyber · 2026-07-20 · major

WordPress pre-auth RCE — found with GPT-5.6 Sol Ultra and $25

Adam Kues at Searchlight Cyber chained a pre-authentication remote code execution in WordPress core — the class of bug exploit brokers pay $500,000 for — after ~10 hours of work and about $25 of GPT-5.6 Sol Ultra API time.

Searchlight Cyber research post card announcing the WordPress pre-auth RCE found with GPT-5.6 Sol Ultra

One researcher, GPT-5.6 Sol Ultra, and $25 of API time chained a pre-auth RCE in WordPress core.

Key specs

Model cost	$25
Wall clock hours	10
Installs at risk	500M+

Quick facts

Researcher	Adam Kues (Searchlight Cyber)
Target	WordPress core (pre-auth RCE chain)
Model used	GPT-5.6 Sol Ultra
API spend	~$25 (half of a $200 weekly plan)
Human time	~10 hours
Impact	500M+ WordPress installs affected
Disclosure	Coordinated; details published after fix

What is it?

The WordPress pre-auth RCE is a full exploit chain against WordPress core — the CMS behind roughly 43% of the public web — found by Adam Kues at Searchlight Cyber using GPT-5.6 Sol Ultra. Brokers on dark-market exploit desks advertise this class of bug at around $500,000; Kues spent about $25 in API calls to build it.

How does it work?

Kues rewrote OpenAI's math-conjecture prompt into a bug-hunt harness that ran up to four Sol Ultra agents in parallel for around six hours, with web search and git history disabled so the model had to reason from source. The agents surfaced a validation-vs-execution desync in the batch API, which anchored a chain that layers SQL injection through author_exclude, cache poisoning, embed abuse, a post-hierarchy cycle gadget, changeset admin escalation and hook-system replay for persistent code execution.

Why does it matter?

Pre-auth RCEs against WordPress core are so rare that private buyers pay six figures for them; this one was surfaced in about ten hours of human work on top of $25 of GPT-5.6 Sol Ultra. It is the clearest public evidence yet that frontier reasoning models are compressing the cost of serious offensive security research, which changes the threat model for every WordPress operator and for defenders who assumed AI-assisted attackers were still a future problem.

Who is it for?

security researchers, WordPress operators, red teams, policy folks tracking AI-and-cyber

Frequently asked questions

How did GPT-5.6 Sol Ultra find a WordPress zero-day?: Adam Kues adapted OpenAI's math-conjecture prompt into a bug-hunt harness that ran up to four Sol Ultra agents in parallel over about six hours, restricted from web search and git history so the model had to reason from source. It surfaced a bypass in WordPress's batch API that anchored the RCE chain.
How much did the WordPress RCE cost to find?: Searchlight Cyber estimates about $25 of GPT-5.6 Sol Ultra API time, roughly half of a $200 weekly subscription. Kues put in about ten hours of human work on top of that — for a bug class that dark-market exploit brokers advertise at $500,000.
What is the WordPress RCE vulnerability chain?: The chain starts with a validation-vs-execution desync in WordPress's batch API, then combines SQL injection via author_exclude, cache poisoning, embed abuse, a post-hierarchy cycle gadget, changeset admin escalation and a hook-system replay. Together they give an unauthenticated attacker code execution on a default install.
Which WordPress sites are affected by the GPT-5.6 exploit?: Searchlight Cyber says the pre-auth chain works against a default WordPress install and estimates 500 million+ sites are exposed. Patches are being coordinated; site operators should apply the current WordPress security release as soon as it lands and audit for the batch-API and embed vectors described in the write-up.
Does this mean AI can now find zero-days on its own?: Not on its own — but close. Adam Kues framed GPT-5.6 Sol Ultra as a force multiplier: parallel agents chewed through code paths a human would take days to trace, and the human still stitched the chain together, judged exploitability and did coordinated disclosure. It is the strongest public demo yet of AI-assisted vulnerability research.

Try it

Read the technical write-up at slcyber.io