Loading summary
Transcript1 lines
- [00:02]
A
Cybersecurity Today would like to thank Meter for their support in bringing you this podcast. Meter delivers a complete networking stack, wired, wireless and cellular in one integrated solution that's built for performance and scale. You can find them at meter.com CST Anonymous Crime Tip System breach exposes informants Google warns Quantum Q Day could arrive by 2029 AI supply chain risk expands, documentation can be poisoned, and GitHub Copilot can use your code to train their AI. This is cybersecurity Today. I'm your host Jim Love. A breach at P3 Global intel, which provides TIP submission systems used by police, government agencies and schools, has revealed that millions of records, including the identities of tipsters and the people they reported, have been stolen. The group claiming responsibility calls itself Internet if Machine, and it says it stole more than 8 million submissions, amounting to roughly 93 gigabytes of data. The leaked database includes deeply sensitive personal information names, email addresses, phone numbers, home addresses, license plate numbers, Social Security numbers, and even criminal histories of individuals named in tips. In some cases, it may be possible to identify the tipsters themselves through the same data, undermining the core promise of anonymity. The scale of use makes this far more serious. P3's clients include the US government agencies such as the Air Force, Army Criminal Investigation Division, Homeland Security Investigations, the Secret Service, the IRS Criminal Investigation Division. Federal records show departments including Defense, Homeland Security, justice and interior paid nearly $1.3 million between 2020 and 2025. But the most striking exposure may be in education. More than 30,000 students have used the system, and the leaked data includes reports from students describing bullying, self harm, suicide threats and and even potential violence. What also makes this breach different is how much of the system itself was exposed. The attackers say they exploited multiple security flaws, including plain text, storage of credentials and misconfigured features, which would contradict the company's marketing about security. They also uncovered an internal page showing that clients could request tipster's IP addresses, raising concerns about how anonymous the system really is. Company materials describe a feature called Session Information Disclosure, which is turned off by default, but when it's enabled, it allows P3 to capture tracking information for up to 90 days and to provide it to clients upon request. The company says this is intended to address misuse of the system, such as threats against life or property. The group behind the breach made their intent clear, and in a message left with the data, they wrote, remember, folks, don't do the dirty work for the pigs. Investigating crime is their job, not yours. They don't care about you. They want convictions and prisoners to fuel the for profit prisons CEO JPD Gilbeau said in a statement to San to this point we have not confirmed that any sensitive information has been accessed or misused. As many of you will already know Q Day the point where quantum computers can break today's encryption has been discussed for years. But now Google is putting a timeline on it, warning that banks, governments and technology providers need to be ready for that moment as early as 2029. That contradicts most estimation, which pushes this off into the 2000-30s or even 2000-40s. But in a recent blog post, Google said that quantum computers would pose a significant threat to current cryptographic standards before the end of the decade and urged organizations to begin preparing. That's earlier than many previous projections, but it comes from a company deeply involved in building quantum systems, particularly in areas like error correction and scaling, which are seen as the key barriers to making these machines practical. As many of you know, Q Day refers to the moment when quantum computers can break widely used public key encryption systems such as RSA or the elliptic curve cryptography, the foundation of secure communications across banking, government and the Internet. There is still debate about how realistic the 2029 timeline is, but the direction is clear. The more immediate issue though, and again something many of you will be familiar with, is is the harvest now, decrypt later? Problem attackers don't need to break encryption today, they just need to collect encrypted data now and wait. When quantum capability arrives, that data becomes readable, so that shifts the conversation from future risk to present day exposure. Organizations with long lived sensitive data, intellectual property, health records, government information are already on the clock. Transitioning to quantum resistant approaches, often referred to as post quantum cryptography is underway with standards emerging from bodies like the National Institute of Standards and Technology, nist. There's been a lot of talk about qday over the years and plenty of hype, but when Google starts putting timelines on it, it might be time to listen. Because the deadline isn't Q Day itself. We have no control over that. What we can control is the last day. The data captured can be in a form that can be decrypted later. It turns out that in the age of AI systems, it's not just the code that we need to inspect and protect, it may be the documentation as well, especially when that documentation is community generated, and especially when it's used for AI coded systems. An article in the Register highlights a new attack vector indirect prompt injection through documentation that AI systems rely on. Instead of attacking code directly, malicious instructions can be embedded in the materials that AI tools use to generate that code Set out to solve a familiar problem. API documentation is often outdated, incomplete or or just wrong. To address that, he created a GitHub based resource designed to provide developers with more accurate, up to date API information and documentation that AI systems could also consume. What he didn't anticipate is that this kind of system could also be used to distribute indirect prompt injections. If AI tools are pulling from these sources, then poisoned or manipulated content could can influence the code they generate not by exploiting the model directly, but by shaping what it reads. We've already seen one version of this problem. AI models have been known to hallucinate package names, and attackers have taken advantage of this by publishing malicious packages under those names. But this approach is more deliberate. Instead of waiting for hallucinations, attackers can seed the documentation layer itself. And this is where it gets uncomfortably familiar. Love hate relationship with documentation, it's often the last thing done, and not many people enjoy writing it. The assumption is usually that the code should tell the story, or that I've put comments in the code. So when someone provides documentation that looks polished seems to work, saves time, how many of us are going to read it end to end, especially looking for something as subtle as a prompt injection? According to the researcher, it's unlikely that this is limited to one platform or one offering. Many systems that make community authored documentation available to AI models fall short when it comes to content sanitization, making this a broader issue across the ecosystem. So the new attack surface isn't just the software supply chain, it's the knowledge layer that AI depends on, and that might prove to be even harder to secure. Finally, a heads up for developers using GitHub copilot changes are coming that could affect how your code is used behind the scenes. According to reporting from Neowin and other outlets, G GitHub plans to use copilot interaction data, including prompts, code snippets and context from Free Pro and Pro plus users to help train and improve its models, unless those users opt out. In some cases, that information could also be shared across systems. For many developers, this raises a familiar concern. If your code is part of the training loop, where does it end up and who else might benefit from it? And there are some important nuances. Enterprise environments may not be subject to the same data Usage policies, and GitHub provides options to limit or turn off certain types of data collection. But those controls need to be understood and actively managed. They're not always the default. So this becomes less of a technical issue and more of a governance issue. Organizations need to be clear about what data is being shared, what's being retained, and what's feeding back into the models that teams rely on. Just another thing to check and to raise awareness of. If one company thinks this is a good idea, others are sure to follow. And that's our show. We'd like to thank Meter for their support in bringing you this podcast. Meter delivers full stack networking infrastructure, wired, wireless and cellular to leading enterprises. Working with their partners, Meter Designs, deploys and manages everything required to get performant, reliable and secure connectivity in a space. They design the hardware, the firmware, build the software, manage deployments and run support. It's a single integrated solution that scales from branch offices, warehouses and large campuses to data centers. Book a demo@meter.com CST that's M E T E R.com CST I'm your host Jim Love. Thanks for listening. It.