Short answer: the PS5 "Relapse" jailbreak and the "Overdrive" game-store app are real and widely reported. The claim that artificial intelligence made the hack possible, calling it the "first known case" of AI accelerating hardware reverse engineering, is not. It traces to a single news-aggregator write-up built on a few X posts, and none of the outlet articles we could read about Overdrive or Relapse mention AI at all. This post lays out what is claimed, what is verified, and how you can check for yourself. It deliberately links to no pirated content and no jailbreak tools.
It follows the pattern of other viral AI-capability claims we have covered, such as what Aaronson, Drake and Buterin actually said about AI breaking cryptography and the claimed-vs-verified breakdown of a data-center drone strike. The method is the same: attribute each claim, check it against primary and independent sources, and say plainly where the trail ends.
TL;DR: claim, evidence, status
| Question | Answer |
|---|---|
| What is Relapse? | A public jailbreak for PS5 and PS5 Pro, reportedly published on GitHub on September 29, 2026, covering firmware 7.00 through 13.60 (secondary reports). |
| What is Overdrive? | A homebrew app for jailbroken PS5s that works like a game store; reportedly posted to GitHub on October 9, 2026, about ten days after Relapse. |
| Is 1,669 games confirmed? | Repeated by several outlets and by the first X posts, but at least one outlet says it is unconfirmed. |
| Was AI used? | Claimed by one aggregator. Not mentioned by the outlets we read, the app's reported feature list, or any statement we found from Sony or the developer. |
| Has Sony responded? | No public comment on Overdrive found. Reports say firmware 14.00 patches the underlying hole. |
| Is it safe? | No. Piracy, ban and malware risks are widely flagged. |
What is being claimed?
The story began on X. International Cyber Digest posted that a homebrew app called Overdrive "turns jailbroken PS5s into a store of 1,669 games," that it reached GitHub on October 9, ten days after the Relapse jailbreak, and that it pulls games from file hosts like MediaFire. Other accounts repeated the figure, and an automated reply account on X summarized the app as an on-console store with a 1,669-title catalog tested on firmware 13.60.
A news-aggregator article published on October 12 then layered on two further claims. First, that the breach "was made possible by using artificial intelligence to reverse-engineer the console's hardware security in a matter of weeks," and that this is "the first known case of AI significantly accelerating the process of hardware reverse engineering." Second, that unnamed market observers think the release could cost game publishers billions in lost revenue, naming Rockstar Games.
Notice what each claim rests on. The AI framing is attributed to one X account describing it as a "clear example" of AI speeding up hardware reverse engineering, which reads as opinion rather than reporting. The billions-in-losses line is attributed to another X user. Neither is a source with access to the developers or to Sony.
What is verified?
A magnifier over a claim bubble marked with a green check, illustrating how to verify the PS5 Relapse jailbreak AI claim
The app and its timing
The outlets that wrote about Overdrive agree on the basics. PBX Science describes it as a store-style interface listing commercial games and DLC with no prices, showing download size and a review score instead. It says that per its GitHub page the app supports DualSense controller navigation, up to 64 simultaneous downloads, on-the-fly archive extraction, and handoff to a game-mounting tool, and that it cannot be installed on an unmodified PS5. It dates the GitHub post to October 9, 2026 and a demo clip on X to October 11.
Khel Now reports a catalog of about 1,669 games and DLC pulled from hosts such as MediaFire, and notes that the free-download claims come from unverified "reports." PBX Science adds a similar but smaller storefront, Orbit Store, listing 147 games linked from Archive.org, whose developer says it hosts no files itself. It also notes mixed reactions: some users reported downloads failing during extraction.
The jailbreak
Relapse is covered by secondary outlets, not by Sony. Coverage syndicated from iPhoneSoft and other sites says it was published on GitHub on September 29, 2026, targets firmware 7.00 through 13.60 on both PS5 and PS5 Pro, and works by chaining a WebKit browser bug with a kernel exploit. Reports credit a developer going by ntfargo, with help from others in the scene. They also warn the exploit can freeze or crash the console and does not survive a full reboot, so it must be rerun each time the console starts.
Those technical details come from secondary reporting and we could not open the exploit's repository from this environment, nor would we link to it. Treat the firmware range and mechanism as reported, not independently tested.
Sony's position
We found no Sony statement about Overdrive. Khel Now notes Sony does not support PS5 jailbreaking or console modification and says unofficial software can violate PlayStation's terms of service. Secondary reports say firmware 14.00 closes the underlying hole and that PS5 has no official downgrade path, so consoles already updated cannot use the exploit. We did not find Sony's own release notes confirming that.
Where does AI enter the picture, and what is the evidence?
This is the gap. Across the Overdrive articles we could read, the word AI does not appear in connection with the app: PBX Science and Khel Now each describe the app, its features and its risks without any mention of AI-assisted development. The jailbreak reports we found describe the exploit chain and its credited developers, again without mentioning AI. Nothing in the app's reported feature list, which is a download manager and catalog browser, requires AI either.
Some further points help calibrate the claim:
- Overdrive is not the hack. The app is a front end that runs on a console already jailbroken by Relapse. Even if AI helped write a storefront, that is software development, not hardware reverse engineering. The "first accelerated hardware hack" framing blurs the two.
- Relapse's described mechanism is a software chain. A browser flaw plus a kernel exploit is a software vulnerability chain. The aggregator's phrase "reverse-engineer the console's hardware security" does not match what secondary reports describe.
- "In a matter of weeks" is not a measurement. No timeline from the developer for how long the research took has been cited. Without that, "AI accelerated it" cannot be evaluated, because there is nothing to compare against.
- "First known case" is a superlative with no support. AI-assisted decompilation and porting is already documented elsewhere, for example in our post on AI-decompiled game ports running in the browser and in coverage of a reverse-engineering MCP server for agents. That does not prove AI was used here, but it shows why "first" needs evidence.
None of this means AI was not used. Developers in the console scene do use coding assistants, and a claim like this could be true. But it has not been stated by the people who would know, so it should be read as unverified.
What the numbers do and do not tell us
The 1,669 figure deserves its own caution. It is a catalog count produced by a pirate app, repeated verbatim across posts with no independent audit, and at least one outlet flags it as unconfirmed. A catalog entry is a link to a file on a third-party host, not proof that a working download exists: users in the reports describe failed extractions. The "billions in lost revenue" line is speculation about a single publisher with no calculation shown, and we do not repeat it as fact.
The same discipline applies to the AI angle. Large numbers and "firsts" travel fast because they are easy to repost. A short checklist helps:
- Who made the claim, and could they plausibly know? A developer, a vendor, or a regulator can. An aggregator summarizing posts cannot.
- Is there a primary artifact, such as release notes, a changelog, a commit message, a write-up from the developer, or a vendor statement?
- Does the claim match the mechanism described elsewhere?
- Is a superlative attached ("first", "billions")? Ask for the comparison.
Why the claim is plausible enough to take seriously
It is worth explaining why the story spread. Over the past year, AI systems have become noticeably better at reading binaries, tracing code paths and proposing vulnerability candidates. Anthropic has launched programs around AI-assisted vulnerability work, including a free open-source scanner for maintainers, and security vendors now pitch autonomous attackers, for example Armadin's AI attack swarms. If AI genuinely speeds up work that used to take a specialist months, consoles, phones and embedded devices are natural targets, because their security depends on secrets staying obscure.
That context makes the PS5 claim feel believable, which is exactly why it should be tested. A believable claim with no source is still unsourced.
What to do if you own a PS5 or run a security team
If you own a PS5:
- Do not install unofficial stores or run exploits. Outlets warn of piracy, account and hardware bans, and malware risk from unofficial software, and reports describe exploit runs that can crash or freeze the console.
- Keep firmware updated through the official system settings. Reports say the update that closes the Relapse hole is firmware 14.00, but confirm in Sony's own release notes.
- Do not download "free game" files shared on social media. Files from third-party hosts can bundle malware, and nothing about a viral screenshot proves a file is safe.
If you build or secure products:
- Assume exploit development cycles may shorten as AI tools improve, and shorten your own patch cadence accordingly.
- Do not rely on obscurity for hardware or firmware secrets. Treat anything shipped on a device as eventually readable.
- If you deploy AI agents with tool access, put a control layer between the agent and dangerous actions. AgentBeam, from the explainx.ai team, stops AI agents before they take risky actions, and our overview of AI agent security platforms compares the options.
How to check an AI-capability claim yourself
A magnifier following a trail of footprints back toward the original source of an AI hacking claim
- Find the first post and see whether it makes the AI claim or someone downstream added it.
- Look for the developer's own write-up, repository README or commit history. Credits, tooling notes and thanks often mention the tools used.
- Read two independent outlets. If neither mentions the claim, it is probably not established.
- Search for the vendor or platform holder's statement. Sony has said nothing about AI in this case.
- Reread any number. A catalog count or a loss estimate with no method is a headline, not data.
We will update this post if the developer, Sony, or a credible outlet confirms or rules out AI involvement. For more on why checking AI output and AI claims costs time, see our look at the verification tax.
Bottom line
Relapse and Overdrive are real and carry real risks for anyone who uses them. The AI claim is a single-source assertion with no primary evidence, and the mechanism reported elsewhere does not obviously need AI. Until the developer or a credible outlet says otherwise, the accurate summary is: a console jailbreak and a pirate storefront appeared in quick succession; whether AI helped is unknown.
Related reading
- Can AI break cryptography? Claimed vs. verified
- Yandex data-center drone strike: claimed vs. verified
- AI-decompiled game ports in the browser
- Reverse-engineering MCP server and agent security
- Anthropic's free AI vulnerability scans for open-source maintainers
- AI agent security platforms compared
- The AI verification tax
Details reflect public reporting as of October 13, 2026 and may change as the developer, Sony or outlets add information.