NB572: Quantum Switches and Flying Cars
Network Break hosts Drew Connery-Murray and guest Scott Robon cover a range of networking and IT news including Cisco's quantum switch prototype, Cato Networks' enterprise browser, Anthropic's Mythos model both finding and introducing vulnerabilities, financial results from Nokia and Intel, a European satellite startup, and a Chinese company's flying car ambitions. Listener follow-up addresses concerns about Anthropic's Project Glasswing potentially exposing proprietary source code and Linux 7.0's removal of legacy protocols. Throughout, both hosts balance optimism about emerging technology with measured skepticism about vendor hype.
Summary
The episode opens with listener follow-up on two prior topics. Frank raised a 'reverse Trojan horse' concern about Anthropic's Project Glasswing, arguing that by inviting major companies like Microsoft, Cisco, and Apple to run the Mythos model against their proprietary codebases to find vulnerabilities, Anthropic could potentially retain and train on that source code. The hosts acknowledged the concern but argued that the legal departments of those large companies would almost certainly have ironclad protections in place, and that Anthropic would be foolish to provoke such well-resourced organizations. A second listener, Stephen, commented on Linux 7.0 removing legacy protocols and drivers including ham radio, ATM, ISDN, and ISA network card support, framing it as a 'Marie Kondo moment' driven partly by AI-generated bug reports. Scott connected this to broader industry patterns of technical debt cleanup, comparing it to major NOS overhauls like IOS to IOS XR and Nokia's SROS to SR Linux transitions.
On the news front, Cisco announced a universal quantum switch prototype capable of routing quantum information between quantum computers using different modalities over existing telecom fiber. It has so far only validated polarization encoding, with other modalities like pulse timing and frequency yet to be confirmed. The hosts were cautiously optimistic, noting Cisco's prior work on entangled photon transmission over NYC fiber, while emphasizing this is still a prototype with no purchasable SKU. Scott highlighted the urgency of quantum-safe cryptography, noting that a single production quantum system could break conventional crypto rapidly.
Cato Networks launched an enterprise browser integrated into its Universal ZTNA platform, available at no extra charge for existing license holders. The browser offers threat inspection, shadow IT discovery, and AI tool access enforcement. Both hosts agreed the primary challenge is user adoption, given how entrenched people are in their existing browser workflows and extensions. Scott suggested a Chrome extension approach might lower the barrier to adoption.
The Anthropic Mythos discussion had two sides. On the positive side, Mozilla reported that Mythos helped find and fix 271 vulnerabilities during a preview evaluation, with Mozilla stating the model is 'every bit as capable' as the world's best human security researchers. On the negative side, Forbes reported that Claude's Opus 4.6 model introduced serious security defects and saw nearly a 50% drop in code quality, with AMD and Veracode also noting Claude writes less secure code than competitors. The hosts attributed the degradation partly to token reduction optimizations. Scott noted this reflects a broader problem: AI model behavior can change unpredictably between versions, undermining workflows built around specific models.
Meta announced layoffs of 10% of its workforce (around 8,000 employees), while Microsoft offered voluntary buyouts to employees at senior director level and below whose age plus years of employment equals 70 or more, affecting up to 7% of staff. The hosts noted the Microsoft formula raised potential age discrimination concerns, though legal review presumably preceded the announcement. Scott reflected personally on the human cost of workforce reductions and praised Microsoft's softer buyout approach relative to outright termination.
Nokia reported Q1 2026 revenues of 4.5 billion euros, up 4% YoY, with its Optical Networks segment growing 20% and AI/cloud customer sales up nearly 50%, validating its 2024 acquisition of Infinera. The company raised its full-year Optical and IP networking growth forecast to 18-20%. Intel posted Q1 2026 revenues of $13.6 billion, up 7% YoY, but reported a $3.7 billion loss, significantly worse than the $877 million loss in the same period last year. Despite this, Intel's stock has risen, partly due to AI-driven CPU demand and perceived government backing making it a 'too important to fail' bet. Scott expressed hope that Intel could develop into a TSMC-style contract foundry to address domestic semiconductor manufacturing gaps.
French satellite startup Univity received 27 million euros to build two demonstration LEO satellites offering broadband and direct-to-mobile 5G. The hosts noted the company is also playing a European digital sovereignty angle as governments seek to reduce dependency on U.S. tech infrastructure. Finally, Chinese EV maker Xpeng showcased a quadcopter-style flying vehicle that docks on the back of one of its electric cars, with deliveries promised for 2027. The hosts were playfully skeptical about calling it a 'flying car' but expressed genuine enthusiasm for the concept.
Key Insights
- Scott argues that AI model behavior changes unpredictably between versions, which is particularly problematic for organizations that have built automation workflows around specific model releases, since continued improvement cannot be assumed.
- A listener argued that Anthropic's Project Glasswing functions as a 'reverse Trojan horse,' where major companies feeding their proprietary codebases into Mythos could inadvertently give Anthropic access to previously unseen source code for potential model training.
- Mozilla's evaluation of Mythos found it capable of identifying 271 vulnerabilities and stated there is 'no category or complexity of vulnerabilities that humans can find that this model can't,' offering one of the strongest third-party endorsements of the model's security research capabilities.
- Forbes reported that Anthropic's Claude Opus 4.6 experienced a nearly 50% drop in code quality and began introducing serious security defects, with the degradation potentially tied to token reduction optimizations that cut costs but compromise output.
- Scott observed that tracking where respected industry professionals are moving on LinkedIn serves as a reliable signal for which emerging technology areas, such as quantum computing, are becoming more substantively real versus being driven by hype.
- Cisco's quantum switch prototype has only validated a single modality (polarization encoding) and cannot yet be purchased, leading Scott to caution against confusing an important research milestone with a commercially deployable product.
- Scott contended that Intel's most strategically valuable contribution would be developing a TSMC-style contract foundry model, which would address critical domestic semiconductor manufacturing capacity gaps rather than simply competing in the consumer CPU market.
- The hosts noted that the Microsoft voluntary buyout formula—age plus years of employment equaling 70 or higher—raises potential age discrimination concerns, and Scott reflected that experienced workers are increasingly aware of and suspicious about such structural workforce reduction mechanisms.
Topics
Full transcript available for MurmurCast members
Sign Up to Access