Security researchers exploited 32 previously unknown vulnerabilities and earned $388,500 on the first day of Pwn2Own Ireland 2026, BleepingComputer reported Tuesday. Targets included AI software such as OpenAI's Codex coding agent and Oracle's Autonomous AI Database, alongside a flagship Samsung phone, printers and smart home devices.

What Happened

The Vietnamese team VinSOC, made up of Vu Chi Thanh and Huynh Duc Tin, topped the day with $80,000. It earned $40,000 for chaining seven zero-days against the Philips Hue Bridge Pro smart lighting hub and another $40,000 for a five-bug chain against Oracle Autonomous AI Database.

OpenAI Codex was compromised with a single argument-injection bug, BleepingComputer reported. Researchers also demonstrated zero-days in LiteLLM, an open-source gateway that routes requests to different large language models.

RelatedSuspected Chinese Spies Exploited BeyondTrust Remote Support Flaw Since August, Google Says

The Samsung Galaxy S26 was hacked twice, in entries involving Interrupt Labs, Ikotas Labs and Nguyen Thanh Dat of Viettel Cyber Security, though some of the bugs used were already known to Samsung. Two multifunction printers, the Lexmark CX532adwe and the Canon imageFORCE 1643F, were exploited, and the Sonos Era 300 speaker fell to a chain of four vulnerabilities after also being compromised at a previous event.

Not every attempt succeeded. White Noise Club could not get its exploit against the Google Pixel 10 working within the time limit.

Why It Matters

The results show AI tooling becoming a mainstream target category next to the phones and routers that have long anchored the contest. Coding agents such as Codex run commands, edit files and call external tools on a developer's behalf, so an argument-injection flaw, in which crafted input is passed through as an extra command-line option, can turn a helpful assistant into a way to run attacker-controlled actions. That a one-bug exploit was enough suggests these products have not yet been hardened the way mature browsers and operating systems have.

LiteLLM and Oracle's database sit deeper in enterprise AI stacks. A gateway like LiteLLM often holds API keys for several model providers, and a database service marketed for AI workloads stores the data those models draw on, so flaws in either can expose more than a single application.

Context

Pwn2Own is run by the Zero Day Initiative, the vulnerability research program owned by Trend Micro, and pays researchers for working exploits against fully patched products demonstrated live. Vendors receive 90 days to fix the flaws before details are disclosed. Last year's Ireland competition paid out $1,024,750 for 73 zero-days, according to BleepingComputer.

What to Watch

The contest runs two more days, with entries covering phones, printers, smart home equipment, AI infrastructure and wellness and health care devices. Running totals will show whether this year's event approaches last year's payout. After that, the focus shifts to patches from OpenAI, Oracle, Samsung, Philips, Sonos and the printer makers within the 90-day window, and to whether AI tool vendors disclose how they addressed the bug classes the contest exposed.