The FBI removed an Accenture contractor on Monday after a data breach exposed sensitive personal details of thousands of bureau employees, Reuters reported. FBI cyber chief Brett Leatherman said the incident “occurred as the result of a security failure of a platform managed by a third-party organization — after a contractor failed to implement a security patch explicitly issued to secure the platform.”
More info here: FBI removes Accenture contractor after ShinyHunters data breach
Jason Brown, Director of Customer Advisory, Counter Fraud Lead, iCOUNTER
“The FBI breach came through a PeopleSoft HR platform that a third party managed. According to the FBI, a contractor failed to apply a patch that Oracle and Google flagged as critical back in June. ShinyHunters claims it got in through that flaw in September. Third-party risk lives in that gap. Organizations hand third parties some of their most sensitive systems and then have very little insight into whether those systems are patched, monitored or exposed. In this case, Reuters reports the result was counterintelligence role descriptions, home addresses of human intelligence operatives, and medical and psychiatric records for bureau employees. Every contract for a managed platform should spell out how quickly critical patches have to be applied, require the third party to prove it was done, and give the customer the right to check. Security teams also need to stop treating third-party managed systems as someone else’s problem. Keep an inventory of every platform a third party runs on your behalf, know what data sits in it, and watch for warning signs, like that platform showing up in exploit activity or criminal forums. Groups like ShinyHunters look for exactly this kind of weak link, because one unpatched platform at a contractor can expose an entire workforce. Intelligence on which of your third parties are being actively targeted gives you a chance to act before the data is gone. Outsourcing the platform leaves the accountability with you. When it’s your employees’ data, the breach is yours.”
ShinyHunters is busy for sure. Thus proving that this wasn’t going to stop them. Thus if I were you, I would harden things up so that ShinyHunters don’t get in at all.
UPDATE; More commentary has come in.
Justin Beals, CEO & Founder, Strike Graph
“The FBI deserves credit for saying plainly what happened. A contractor failed to apply a patch that was explicitly issued to secure the platform, and the bureau named that as the cause and removed the contractor. Most organizations would bury that finding. The harder truth is that the FBI’s workforce data was only as secure as the patch cadence of a team it didn’t directly manage.
Federal contracting keeps running into this gap. Agencies and primes collect attestations that controls are in place, but an attestation only tells you what a vendor intended to do. It doesn’t tell you whether the patch went in after Google warned that ShinyHunters was going after vulnerable PeopleSoft instances. A DHS prime contractor I work with put it to me directly when we were talking about his obligations as an affirming official: ‘I don’t think a security questionnaire is going to cut it.’ He’s right. A questionnaire is self-attested and unverifiable, and he’s signing for every subcontractor below him.
Expect this to show up in enforcement. DOJ recovered $52 million under its Civil Cyber-Fraud Initiative in FY2025, and those cases turn on the gap between what a contractor said about its controls and what was actually running. Removing the contractor was the right call, but it happened after the damage was done. The agencies and primes that avoid the next headline will be the ones requiring continuous evidence from their suppliers, including time-to-patch on critical systems, rather than a form that gets filed once and never revisited.”
Joe Brinkley, Director of Offensive Security Research & Community, Cobalt
“Outsourcing the management of a platform doesn’t outsource the risk that comes with it. Attackers understand this well, which is why groups like ShinyHunters focus on widely deployed enterprise applications where responsibility for patching sits somewhere between the customer and the service provider. When ownership is split, critical updates tend to fall into the gap, and that gap is where real-world exploitation happens. The approach also scales well for attackers: once they have a working method against one platform, they can sweep the internet for every instance that’s lagging behind, regardless of who manages it. Organizations need to treat third-party-managed systems as part of their own attack surface. That means setting clear contractual ownership of remediation timelines and independently validating that patches were actually applied, because a status report is not evidence that the exposure is gone. The organizations that hold up best test their trusted relationships the same way an attacker would, rather than assuming someone else has it covered.”
UPDATE #2: Kevin Surace, CEO, Token said this:
“What makes this incident so troubling is that this was not some mysterious new zero day attack. It was a well known, critical vulnerability rated 9.8 out of 10, which essentially means patch it urgently, and Oracle had already issued the fix. Accenture reportedly failed to implement that patch, leaving the vulnerable system exposed. ShinyHunters was likely scanning large numbers of systems looking for exactly this situation: an internet facing endpoint that remained unpatched. They found one. The attackers did not need to invent a sophisticated new technique; they simply exploited a known vulnerability that should have already been closed.
“This was an entirely preventable breach and a stark reminder that basic patch management remains one of the most important disciplines in cybersecurity.”
Australia weighs AI copyright changes as AI giants push for greater access to training data
Posted in Commentary with tags Anthropic, OpenAI on October 6, 2026 by itnerdAustralia is weighing new rules governing AI as executives from Anthropic and OpenAI appeared before a parliamentary inquiry Tuesday to address AI safety, cybersecurity incident reporting, model training and the country’s existing laws, according to ABC News.
Both Anthropic and OpenAI said they would support laws requiring AI companies to report breaches carried out by their agents. OpenAI Chief Strategy Officer Jason Kwon said a legal framework could establish a standard for when incidents must be disclosed rather than leaving those decisions solely to AI companies. Anthropic’s Australian policy chief David Masters similarly said the company would be open to mandatory disclosure requirements.
The hearing also examined how Australia’s existing copyright laws apply to AI training. Anthropic told lawmakers that requiring licenses for every piece of online content used to train models would be technically unworkable, while Australian media and creative-industry representatives opposed proposals that could allow AI companies to use material unless rights holders specifically opt out.
OpenAI acknowledged at Tuesday’s hearing that its internal escalation and notification process should have been better. The parliamentary inquiry is holding hearings through October 9 and is expected to issue its final report by November 30.
Jacob Krell, Senior Director: Secure AI Solutions & Cybersecurity, Suzu Labs:
“Australia’s AI hearing links security and copyright through one accountability problem. Creators and affected organizations are being asked to absorb the cost of AI companies operating at scale.
“A proposed opt-out copyright model would make rights holders police whether their work enters model training. With incident reporting left to company discretion, affected organizations can wait while a provider decides whether an agent’s unauthorized access deserves disclosure.
“OpenAI’s agent accessed Australia’s Medicare statistics reporting service, and the company took three months to notify the government. OpenAI Chief Strategy Officer Jason Kwon said the company should have handled the response better, while both OpenAI and Anthropic backed mandatory disclosure laws. Mandatory reporting works when the clock starts at the first unauthorized tool call. In my work with agentic systems, authorization is the control point. The provider can investigate whether data was copied while the regulator already knows an incident occurred.
“Copyright needs the same discipline. Anthropic says licensing every piece of online content would be technically unworkable, while the Australian Broadcasting Corporation’s Kate Gilchrist told the inquiry an opt-out model puts the burden on rights holders. The answer to a licensing problem cannot be asking writers, publishers, musicians, and developers to patrol the internet for every training use.
“Australia should require the companies controlling the training pipeline and agent runtime to carry the cost of consent, evidence preservation, and disclosure.”
ㅤ
Ryan McCurdy, VP of Marketing, Liquibase:
“As AI agents start taking actions across more systems, incident reporting is going to depend on having a clear record of what they actually did.
“Knowing an agent accessed a system isn’t enough. You need to know what it changed, what policies it passed, what authorized the action, and what actually happened as a result. That evidence needs to be created as the agent works. If something goes wrong, you shouldn’t have to piece together the agent’s actions after the fact.
“The more authority we give AI to act on its own, the more important that trail becomes.”
Australia has the right idea here. And knowing them, they are also willing to apply strict enforcement as well. The question is if OpenAI and Anthropic will listen and take the right action so as to not fall victim to this law.
Leave a comment »