The IT Nerd

Meta launched Muse Glimmer, a new open-weight AI model designed to perform agentic tasks directly on a Mac or PC using a single graphics card.

CEO Mark Zuckerberg also said Meta plans to release the weights for Muse Spark 1.2, its most advanced model, and called for U.S. policies that make it easier for American companies to compete with Chinese open-weight AI developers.

Reports noted that Hugging Face recently used a Chinese open-weight model to defend against an attack by a rogue OpenAI model because closed models restricted certain cybersecurity uses. Zuckerberg also said Meta will establish a governance structure giving independent directors authority to approve safety criteria for future model releases.

Doc McConnell, Head of Policy and Compliance, Finite State:

   “Today’s essay from Meta on a “Positive AI Future” focuses on the theoretical benefits of widespread, accessible AI for the public. The cybersecurity industry is primed to be receptive to this message after hearing the same lofty promises all last week during the Black Hat and DEF CON conferences in Las Vegas. But to achieve this positive future, society needs a way to select the best model, not the cheapest, or the first-to-market, but the one that users can rely on.

   “As companies race to produce competitive AI, the market will ultimately reward those that provide the greatest transparency into how they’re built, how they work, and how they stay safe and secure.

   “The AI labs are expecting users to trust their products with their most sensitive information: personal preferences, medical history, customer data. The labs need to earn that trust.”

Meta hasn’t earned my trust. And I expect that others feel that way. The question is can they earn that trust back and keep it.

Remember the AI image generator that Meta has force fed Instagram users? And the fact that I wrote how to disable it which showed how much of a privacy nightmare it was? Well, I have some good news for you. Meta has had a change of heart based on this statement:

“Our intent was to provide a useful creative tool and to give people control over whether their public ​content could be referenced in this way,”  

And:

“We’ve ​heard the feedback that this feature missed the mark, so ⁠it’s no longer available,” 

Now no longer being available doesn’t make it more private. It just means it’s no longer available for free to the majority of users. If someone is willing to pay enough, privacy won’t be a limiting factor. Keep that in mind as I don’t think we’ve seen the last of this feature.

Meta has launched its own AI image model integrated into the Instagram app. Now Meta and Instagram in the same sentence should send chills down your spine as the trust level of these two has to be zero. And you’d be right. As it stands by default, anyone can use your profile photo and posts as prompts for generative images.

Whisky, Tango, Foxtrot.

I strongly recommend users flip two settings to off to avoid what is guaranteed to be a privacy nightmare. Inside the Instagram app, tap your photo bottom-right then the hamburger menu icon top-right. Scroll down to Sharing and Reuse. There’s a section headed “Allow people to create with and reuse your content.” There are separate toggles for Posts and Reels, and you’ll want to flip both of them off.

The fact that these settings are buried deep within the Instagram app, and are opt out rather than being opt in is very problematic to say the least. But that is how Meta rolls. They want to grab as much data of yours in order to find a way to make a buck or two off of it. Assuming that they can’t leverage it themselves to advance their rather twisted needs and desires. My advice is to simply skip Meta products entirely like my wife and I have. But if you must use them for whatever reason, you have one more privacy item to keep in your tool belt.

As first reported by Wired on Friday, Meta has paused all work with AI data startup Mercor following a confirmed security breach linked to a supply chain attack involving the LiteLLM open-source project, which impacted thousands of organizations globally.

Mercor, which provides proprietary training data to major AI companies including Meta, OpenAI, and Anthropic, said it was among those affected and has launched an investigation with third-party forensic experts.

The breach raised concerns about potential exposure of sensitive AI training data and internal datasets, which are used to develop and fine-tune large language models. Reports indicate that Mercor’s systems were impacted as part of a broader compromise involving malicious updates to widely used AI tooling, though it remains unclear what specific data was accessed.

Michael Bell, Founder & CEO, Suzu Labs had this comment:

   “The Mercor breach is what happens when the companies building the most valuable AI models in the world outsource the creation of their training data to vendors running on Airtable and shared passwords. A single poisoned open-source package gave attackers VPN credentials, and from there they walked through Mercor’s systems and took 4TB of proprietary datasets, source code, and contractor PII.

   “We’ve been investigating these AI data vendors for months and found the same structural failures at Sama, Teleperformance, Scale AI, and Cognizant we see unrotated credentials, info-stealer infections on contractor endpoints, and access controls that don’t exist. The training data behind every major frontier model is sitting inside vendors that wouldn’t pass a basic security audit, and now that data is on an extortion site. This is a national security problem dressed up as a vendor management failure.”

Lydia Zhang, President & Co-Founder,Ridge Security Technology Inc. adds this comment:

   “This incident alerts us that AI training data should be treated as critical infrastructure, subject to stricter security scrutiny and regulation.

   “The breach also underscores the risks of relying directly on open-source projects in enterprise environments. Supply chain attacks, like the compromised LiteLLM library in this case, can introduce vulnerabilities at scale and expose highly sensitive data. 

   “At a minimum, enterprises should adopt thoroughly tested and commercially supported versions of such components, with stronger security guarantees and accountability.”

Noelle Murata, Sr. Security Engineer, Xcape, Inc. provided this comment:

   “Meta’s indefinite suspension of its partnership with Mercor underscores how the AI industry’s rush to outsource training data has effectively liquidated billions in proprietary methodology. By allowing a poisoned version of the LiteLLM gateway (versions 1.82.7 and 1.82.8) to persist in their environment, Mercor gifted attackers 4 TB of data, including the precise “secret sauce” protocols Meta and OpenAI use to tune their models.

   “This was not a sophisticated zero-day; it was a basic supply chain failure where a compromised security scanner (Trivy) was used to poison a niche dependency that nobody bothered to pin. For anyone surprised that an autonomous, interconnected AI stack would eventually expose sensitive data to the internet, the lesson is clear. 

   “If you are not auditing your data vendors for basic dependency hygiene, your IP is already public property. Defenders must immediately scan for litellm_init.pth files, which provide stealthy persistence on every Python startup, and rotate all LLM provider API keys and cloud tokens. Protecting training integrity now requires treating every AI data broker as a high-risk production endpoint and enforcing strict, pinned Software Bill of Materials (SBOM) standards.

   “If your AI supply chain is this leaky, you are not training a model; you are just broadcasting a technical manual to Lapsus$.”

Supply chain vulnerabilities are real. If your organization doesn’t take them seriously, your organization will get pwned. It’s as simple as that. And you can double that if AI is involved.

A recent incident at Meta shows how an AI agent provided guidance that led an engineer to unintentionally expose a large amount of sensitive internal data to employees for a short period of time.

While Meta confirmed the issue was contained and no external data was mishandled, the episode highlights a broader risk as AI agents become embedded in engineering workflows. These systems aren’t just generating suggestions, they’re influencing real actions inside environments that handle sensitive data.

Gidi Cohen, CEO & Co-founder, Bonfy.AI

“Meta’s incident is exactly what happens when you let agents loose on sensitive data without any real data-centric guardrails. This wasn’t some exotic AGI failure, it was a very simple pattern: an engineer asked an internal agent for help, the agent produced a “reasonable” plan, and that plan quietly exposed a huge amount of internal and user data to people who were never supposed to see it.

The problem is that neither the engineer nor the agent had any persistent notion of “who actually should see this data” beyond whatever happened to sit in a narrow context window at that moment. Traditional controls don’t help much here. Endpoint DLP, CASB, browser controls, even basic role-based permissions, none of them are watching the actual content as it moves through an agent’s reasoning steps and tool calls, especially when the agent is running as a system service in some framework.

Our view is simple: treat agents like very fast, very forgetful junior interns and make the data security layer smart enough to compensate. That means three things: constrain what data is even available to the agent via contextual labeling and grounding; give the agent a Bonfy MCP tool it can call inline to ask “is this safe to use or send in this context?” before it takes an action; and inspect what ultimately comes out of the workflow before it lands in email, chat, dashboards, or internal portals. In a Meta-style scenario, those controls would have either prevented the broad internal exposure entirely or at least shrunk the blast radius to something manageable.

As organizations “experiment at scale” with agents, the only sustainable path is to make agents first-class entities in the risk model and put the intelligence where it belongs: on the data that’s being read, composed, and shared, not just on the configuration screens of yet another AI tool.”

The thing is that when you expose AI anything to sensitive data, it can get out there. Samsung banned AI usage for that reason. Keep that in mind if you’re an organization that uses AI

People often ask my why I refuse to have a Facebook account. Or why I killed my Instagram account a few years ago. As well as why I haven’t got onto Threads. The answer is pretty simple. Meta, the company that owns all of those platforms are pretty evil and simply can’t be trusted. Here’s today’s example of why they can’t be trusted:

In 2016, Facebook launched a secret project designed to intercept and decrypt the network traffic between people using Snapchat’s app and its servers. The goal was to understand users’ behavior and help Facebook compete with Snapchat, according to newly unsealed court documents. Facebook called this “Project Ghostbusters,” in a clear reference to Snapchat’s ghost-like logo.

On Tuesday, a federal court in California released new documents discovered as part of the class action lawsuit between consumers and Meta, Facebook’s parent company.

The newly released documents reveal how Meta tried to gain a competitive advantage over its competitors, including Snapchat and later Amazon and YouTube, by analyzing the network traffic of how its users were interacting with Meta’s competitors. Given these apps’ use of encryption, Facebook needed to develop special technology to get around it.

One of the documents details Facebook’s Project Ghostbusters. The project was part of the company’s In-App Action Panel (IAPP) program, which used a technique for “intercepting and decrypting” encrypted app traffic from users of Snapchat, and later from users of YouTube and Amazon, the consumers’ lawyers wrote in the document.

The document includes internal Facebook emails discussing the project.

“Whenever someone asks a question about Snapchat, the answer is usually that because their traffic is encrypted we have no analytics about them,” Meta chief executive Mark Zuckerberg wrote in an email dated June 9, 2016, which was published as part of the lawsuit. “Given how quickly they’re growing, it seems important to figure out a new way to get reliable analytics about them. Perhaps we need to do panels or write custom software. You should figure out how to do this.”

Facebook’s engineers solution was to use Onavo, a VPN-like service that Facebook acquired in 2013. In 2019, Facebook shut down Onavo after a TechCrunch investigation revealed that Facebook had been secretly paying teenagers to use Onavo so the company could access all of their web activity.

If some of that sounds familiar, it should as I’ve written about Onavo before. But here’s the bottom line. Even for Meta, that’s a new low. And it illustrates how untrustworthy Mark Zuckerberg and Meta are. They clearly will stop at nothing to grab as much information about you as they can so that they can find new ways to make money. I for one refuse to be the product. Thus you will not see me use a Meta product for that reason. Meta is a company that needs some government intervention in the US because it’s clear from this example and others that they will not alter their behaviour unless they are forced to.

Business Today is reporting Meta files lawsuit against former VP over alleged data breach: Report. The complaint, filed on February 29 in California state court in Contra Costa County, asserts that Khurana illicitly transferred these documents to his personal Google Drive and Dropbox accounts just prior to his departure from Meta. Here’s the news brief:

Meta has initiated legal action against one of its former vice presidents, accusing him of a ”stunning” act of betrayal after he defected to an undisclosed AI cloud computing startup, as reported by Bloomberg.

Dipinder Singh Khurana, also known as T.S. Khurana, had been at Meta for 12 years, ascending to a senior position as VP of infrastructure. However, Meta alleges that Khurana breached his contract by absconding with a cache of proprietary, highly sensitive, and confidential documents relating to Meta’s business operations and personnel.

The complaint, filed on February 29 in California state court in Contra Costa County, asserts that Khurana illicitly transferred these documents to his personal Google Drive and Dropbox accounts just prior to his departure from Meta. Meta contends that Khurana’s actions were not only disloyal but also had tangible repercussions, as at least eight employees listed on the pilfered documents subsequently left Meta to join Khurana’s new venture last year.

“Khurana’s conduct while leaving Meta, and since then, reflects an utter disregard for his contractual and legal obligations,” the lawsuit states. A spokesperson for Meta told Bloomberg that the company ”takes this kind of egregious misconduct seriously” and will ”continue working to protect confidential business and employee information.”

Troy Batterberry, CEO, EchoMark had this comment:

   “Insider threats represent a significant and growing challenge for organizations, as made clear by the recent case involving an accused former Meta executive. The taking of confidential and proprietary information is not an uncommon situation, especially within enterprises and adequately protecting proprietary information and intellectual property is a prominent issue. The stark reality of what businesses face today regarding data security and insider threats highlights the need to safeguard sensitive information against unauthorized transfers. 

   “Whether or not there’s a breach of legal contract, the misconduct is a blatant misuse of privileged access and a breach of professional expectations and conduct. Actions that involve unauthorized sharing of sensitive information not only breach trust but also undermine the very foundation of an organization’s integrity and security. Incidents like these are not isolated and can have far-reaching consequences for any organization.”

Threats to your organization don’t come from some threat actor in China or Russia. They come from people you trust inside your organization. Thus you need to do everything possible to keep threats from outside and inside from affecting your business.

If you’re trying to get access to Facebook, Instagram or Messenger, good luck with that as all three services appear to be down based on downdetector.ca:

As it stands, you might not be able to log into any of these services, or you might have been forcefully logged out of those platforms. At this time there’s no ETA as to when this will be resolved.

I am also tracking other outages that might be happening at the moment and I will post a separate story once I confirm or deny that those services are working or not.