New realities of an AI-driven world, from emerging cybersecurity threats and attack surfaces to hidden signals in training data.
This edition of the IT news and research roundup has a clear AI-emphasis. Because let’s face it: right now, it’s difficult to think about technology without also rethinking it through the lens of AI. With that in mind, let’s get into the stories and research shaping conversations this month.
Earlier this year, cybersecurity was rocked by Anthropic’s Mythos, a new model with unprecedented ability to find vulnerabilities. Over at The Verge, they look at the post-Mythos landscape, and how AI promises to super-charge the work of no-skill hackers, so-called ‘script kiddies.’
The article offers a timely look at our AI-powered security landscape, along with the threat of what security consultant Katie Moussouris has called a “patchpocalypse”, as companies everywhere scramble to eliminate vulnerabilities.
Bankrupted US airline Sprint Airlines has sold its data to Google for just USD $10 million. It’s thought the anonymized data—including employee communications, and passenger records––will be used to train aviation-based AI technologies.
If this shows anything, it’s the increasingly extreme lengths AI companies are going to do acquire more data.
Read more: Google buys Spirit Airlines data for AI training for just $10 million ↗
Researchers at security firm Varonis have demonstrated a one-click attack against Microsoft 365 Copilot that could access and exfiltrate information from a user’s environment. How did it do this? By asking the AI assistant directly. Although it initially refused, it eventually revealed the guardrails the researchers needed to bypass.
Microsoft has since mitigated the vulnerability, but the incident highlights a broader challenge for IT leaders: as AI assistants gain access to more tools, security risks increasingly shift from the model itself to the systems and permissions surrounding it.
Read more: Microsoft Copilot reveals secret input that allowed it to be hacked ↗
AI safety researchers have published a study on the idea of ‘subliminal learning’, showing how data produced by one model can pass on behavioral traits to another via hidden signals. Even when the data has nothing to do with those traits.
The outcome, as Owain Evans, one of the researchers, said, was: “Datasets consisting only of 3-digit numbers can transmit a love for owls, or evil tendencies”. An amusing story––if it didn’t also contain worrying implications for LLMs (as well as a world) increasingly founded on AI-generated content.
Lastly, another interesting study about LLMs, which in this case created an LLM (LittleLearner) trained to mimic the capabilities of an average 10-year-old.
The study found that injecting new knowledge through post-training and in-context learning did not make the model smarter. Instead, it only "let LittleLearner better utilize existing knowledge", which might be an uncomfortable idea for those convinced that those same strategies might meaningfully contribute to artificial general intelligence (AGI).
Read more: LittleLearner: Language Models Under Pedagogically Controlled Knowledge Exposure ↗
Whether you're managing security, data, or digital transformation, the technologies shaping tomorrow’s workplace continue to raise a whole lot of questions alongside opportunities. Until next time!
Senior Content Marketing Manager at TeamViewer
—
Berlin-based Rebecca is the editor of the Insights section. With extensive experience in long-form writing and editing, she loves nothing more than translating TeamViewer expertise into useful and engaging content.
