This Week in Tech: AI Regulation Shakeups, Tesla's Voice-Controlled FSD, and Solid-State Batteries

1. Anthropic, the White House, and the Mythos 5 Standoff

On Friday afternoon, the Trump administration issued an export-control directive demanding Anthropic suspend access to its Mythos 5 and Fable 5 models to any foreign national—including non-US Anthropic staff. The order came with a 90-minute deadline and the implicit threat of Commerce Department intervention, setting the stage for a rare weekend of direct negotiation between a frontier AI company and the White House. Anthropic CEO Dario Amodei spoke directly with Treasury Secretary Scott Bessent, Commerce Secretary Howard Lutnick, and National Cyber Director Sean Cairncross in a marathon series of calls as the company scrambled to keep the model available. Senior employees including head of safeguards Dave Orr and frontier red-team lead Logan Graham were dispatched to Washington for in-person meetings over the weekend. The company ultimately agreed to restrict access while it negotiated the terms of what it called a 'narrow, non-universal' jailbreak issue.

What Are Fable 5 and Mythos 5?

Anthropic built both models on the same foundation as its Mythos Preview, a system the company once described as too powerful for public release. Fable 5 shipped with safeguards designed to block high-risk responses in cybersecurity and biology; Mythos 5 is the same model but with those guardrails partially lifted. Pricing is set at $10 per million input tokens and $50 per million output tokens—double Opus 4.8 rates, though half the cost of Mythos Preview access. The company reported that 95 percent of Fable 5 sessions ran entirely on Fable responses without falling back to the older Opus 4.8 model, a signal that the new architecture is stable enough for broad enterprise use. The safeguards cover specific categories where advanced AI can cause harm: Fable 5 refuses to generate detailed instructions for exploiting software vulnerabilities or engineering pathogens, and when it detects a request in those domains it redirects the conversation or declines with a generic response.

What Triggered the Action?

A report surfaced showing a jailbreak that could bypass Fable 5 safeguards in at least one configuration. Amazon CEO Andy Jassy subsequently flagged concerns after Amazon researchers red-teamed the model and found edge cases where the safety system could be circumvented. Anthropic counters that the jailbreak was narrow, non-universal, and achievable with existing models like OpenAI's GPT-5.5, meaning restricting Fable 5 alone does not meaningfully reduce risk. Semafor reported that the US government was concerned a China-linked group had accessed the technology, tracing the concern back to a global telecommunications company that had initially been cleared for Mythos Preview access and was later revoked at the government's request. David Sacks, the former AI and crypto czar, confirmed that a 'highly credible trusted partner of both Anthropic and the USG' came forward with the jailbreak report.

The Wider AI Safety Context

Anthropic's warnings about Mythos-class models have been central to its brand since the Preview launch, and the company's dire language about the dangers of releasing powerful models without safeguards created expectations that the government appears to have acted on. But some independent red-teamers have said they were impressed with the level of protections in Fable 5 and that the safeguards were so aggressive they became a source of humor in the cybersecurity community on launch day. Alexis Stamos, now chief product officer at Corridor, told The Verge he organized a public letter signed by dozens of tech and security executives calling for the restrictions to be repealed. 'We're in a race, and I think policymakers don't understand that,' Stamos said. 'There's this weird arrogance, this idea that American labs are hugely ahead of our adversaries that will always be true, that it's really important to restrict access because of that. I just think that's foolish.' The letter argued that if AI regulation is going to happen, it should be rooted in scientific evaluations developed with input from industry and academia, not executive orders based on single red-team reports.

Market and Geopolitical Impact

The standoff has already reshaped how AI companies think about political risk. Stamos said the industry is seeing a wave of backup contracts being signed with non-US companies and open-weight models being deployed on alternative hardware arrangements. OpenAI, Google, and Microsoft have all released comparable cybersecurity-focused products in recent months, meaning the US government could eventually face pressure to regulate all of them or none of them. Cybersecurity leaders have warned that sidelining Mythos 5 could give China a meaningful advantage in finding and fixing software vulnerabilities, a capability that matters for national security as much as any weapons system. The incident has also galvanized international calls for sovereign AI alternatives to American cloud providers, with European and Asian governments accelerating their own AI infrastructure investments in response to what they see as politically motivated US restrictions. For enterprises that had planned to evaluate Mythos 5 and Fable 5 for security operations, the shutdown creates immediate uncertainty and may push them toward alternative platforms sooner than they intended.

2. AI Talent Reshuffle: Noam Shazeer to OpenAI

A Quiet Mega-Hire

Noam Shazeer—co-lead of Google's Gemini project—joined OpenAI this week in one of the most significant talent moves of the year. Shazeer spent roughly twenty years at Google, departing in 2021 to co-found Character.AI, a conversational AI startup focused on personalized chatbot experiences. Alphabet paid an estimated $2.7 billion to acquire Character.AI and bring Shazeer and his research team back into the Google fold in 2024. That reunion lasted less than two years. Shazeer's second exit for OpenAI is a reminder that the AI talent market remains intensely competitive and that even the most prestigious research institutions cannot always hold onto their top researchers when the gravitational pull of OpenAI's compute infrastructure, research culture, and product velocity is strong enough.

What Shazeer Brings to OpenAI

Shazeer has deep expertise in transformer architectures, large-scale language model training, and the product design of conversational AI systems. He was one of the original authors of the 'Attention Is All You Need' paper and has been involved in Google's most sensitive model development work for more than a decade. At OpenAI, he is expected to contribute to both foundational research and product development, potentially accelerating the company's work on agent systems and multimodal capabilities. His hiring also sends a signal to the broader AI community: OpenAI remains the most attractive destination for top AI researchers who want to work on models that are deployed at scale and shape the trajectory of the industry. For enterprises evaluating API providers, Shazeer's move should be factored into any long-term roadmap planning. OpenAI's talent depth is one of its strongest moats, and additions of this caliber reinforce that advantage.

3. Tesla FSD: Voice, Parking, and the Autonomy Gap

New Features in Detail

Elon Musk announced that Tesla's Full Self-Driving system will soon learn parking preferences and accept natural-language commands via Grok voice control. Users could eventually say 'turn right here' or 'drop us off here' and have the car comply, similar to directing an Uber driver or giving directions to a human chauffeur. The parking feature addresses what Musk described as the single biggest reason people currently intervene with FSD: when the car enters a parking lot, it tends to grab the first open spot it detects rather than choosing one based on past driver behavior, proximity to the entrance, or weather considerations like avoiding shaded or damp areas. The Grok voice integration also reflects Tesla's broader strategy of embedding its AI models directly into vehicle software rather than relying on third-party navigation or assistant systems.

The Engineering Reality

Even with these improvements, Tesla's own metrics show FSD still requires driver intervention roughly every 3,000 miles. Crowdsourced data puts critical safety interventions at that frequency, which is predictable and acceptable for a supervised driver-assistance system but is fundamentally incompatible with the 'Full Self-Driving' branding that has been on sale since 2016. Musk claimed that 'critical safety interventions are extremely rare,' though he did not share raw data to quantify that claim. The company also confirmed that it recently unified the AI models powering consumer FSD, the Robotaxi fleet, and Actually Smart Summon into a single architecture in the v14.3.3 update, which improved response consistency and cut summon speeds by roughly 33 percent. Meanwhile, Tesla's unsupervised Robotaxi service in Austin has expanded to cover the entire metro area with only about 20 vehicles—a year after launch—underscoring the vast gap between prototype scale and actual commercial deployment.

Regulatory Pressure

The European Union is also scrutinizing Tesla's FSD claims, with Sweden's transport authority recommending that the bloc vote against approving Tesla's Full Self-Driving Supervised system unless the feature's ability to exceed posted speed limits is removed. The recommendation, sent in a previously unreported April 30 letter to the EU's Technical Committee on Motor Vehicles, lands just before the committee is due to take up the matter on June 30 ahead of an eventual vote on a bloc-wide rollout. Sweden's concern is not about the technology's capabilities but about its behavior: Tesla's system has been observed speeding in testing, and the Swedish authority wants that behavior corrected before it is allowed on European roads. This represents a concrete regulatory hurdle that could delay or block EU rollout entirely, and it underscores that Tesla's self-driving program faces scrutiny on multiple fronts simultaneously—not just from consumers and investors, but from regulators who are increasingly willing to act on safety concerns.

Why This Matters

Parking preferences and voice commands are genuine quality-of-life enhancements that will likely improve the daily driving experience for FSD subscribers. They represent the kind of polish that comes from a mature product and demonstrate that Tesla's neural network approach is yielding measurable improvements in common scenarios. They are not, however, steps toward unsupervised autonomy. The gap between marketing language and engineering reality remains the central design challenge for Tesla's autonomous program. When the headline feature is 'the car will remember where you like to park so you stop having to take over,' the company is describing a sophisticated driver assistance system, not an autonomous vehicle. Reducing interventions is not the same as eliminating the driver, and 'extremely rare' safety interventions are still safety interventions. For investors and customers, the distinction matters enormously: supervised systems carry different liability profiles, different regulatory requirements, and different pricing power than truly unsupervised autonomous vehicles.

4. Honda and QuantumScape: Commercializing Solid-State Batteries

The Technology and Its History

Honda and QuantumScape announced a partnership to commercialize solid-state batteries for electric vehicles. Solid-state cells—which replace the flammable liquid electrolytes used in conventional lithium-ion batteries with solid ceramic or glass-like materials—have been hyped as a battery breakthrough for more than a decade. The chemistry offers the theoretical promise of higher energy density, faster charging, longer cycle life, and dramatically improved safety because the solid electrolyte is non-flammable. Those advantages have always existed on paper; the challenge has been manufacturing cells that are reliable, affordable, and consistent enough for mass-market vehicles. QuantumScape has been one of the most visible players in the space, going public through a SPAC in 2020 and steadily releasing testing data showing progress on cycle life, charge rates, and temperature performance. Honda is one of the largest and most respected automotive engineers in the world, and its involvement adds manufacturing credibility that many battery startups have struggled to establish.

Why Now

Automakers and suppliers are converging on next-generation cell architectures because liquid-electrolyte lithium-ion batteries have largely matured. Incremental improvements to cathode chemistry and cell design are delivering single-digit percentage gains in energy density each year, which is no longer enough to sustain competitive EV range claims as vehicle weight targets push higher and aerodynamics improve. Solid-state chemistry is the clearest near-term path to a meaningful step-change in battery performance—potentially doubling the amount of energy stored in the same physical footprint while offering dramatically faster charging without degradation. Honda's partnership with QuantumScape suggests that at least one major OEM now believes the technology is close enough to factory viability to commit capital and engineering resources. The partnership will focus on validating QuantumScape's cell designs for automotive environments, including temperature cycling across climates from Scandinavian winter to Middle Eastern summer, vibration testing for rough roads, crash safety certification, and real-world degradation testing under repeated fast-charging conditions that actual drivers would impose on daily use.

Market Impact and Competitive Landscape

If Honda and QuantumScape can demonstrate volume production, OEM pricing power could shift toward partners who own proprietary solid-state intellectual property. Investors should watch for pilot-vehicle timelines and cost-per-kWh disclosures in coming quarters, because the economics of solid-state cells will determine whether they remain a premium technology used only in high-end models or become the new standard across the entire industry. The broader EV market is watching closely: Tesla has publicly discussed next-gen cell architectures, BYD continues to refine its Blade Battery platform with improved energy density, and Toyota has long signaled its own solid-state ambitions with aggressive target dates for commercial deployment. A credible Honda-QuantumScape roadmap could accelerate investment across the entire supply chain, creating opportunities for materials suppliers, equipment manufacturers, and charging infrastructure providers. It also underscores a broader industry trend: the next decade of EV competition will be decided as much by battery chemistry as by vehicle design, autonomy software, or brand perception.

5. Hyundai Ioniq 9 Achieves Australia's First ISO 15118-20 V2G

The Technical Milestone

Hyundai completed Australia's first Vehicle-to-Grid discharge using the ISO 15118-20 standard, the second-generation international specification for bidirectional power transfer between electric vehicles and charging equipment. The test paired a Hyundai Ioniq 9 with a StarCharge Halo 7.4 kW bidirectional charger that is Clean Energy Council approved and compliant with the AS/NZS 4777.2 home battery standard. During the discharge test, the Ioniq 9 successfully sent power from its battery back into the grid through the bidirectional charger, confirming that the car, charger, and grid interface all communicate correctly under real operating conditions using the ISO 15118-20 protocol. The test also demonstrated that the bidirectional transfer does not interfere with the vehicle's battery management system, addressing one of the earliest concerns that V2G could degrade battery health over time.

Why ISO 15118-20 Is Significant

The ISO 15118-20 protocol represents a substantial upgrade over earlier V2G specifications. It provides a secure, standardized communication framework that covers authentication, billing, grid management signals, and safety interlocks in a single specification. Earlier V2G implementations required proprietary communication protocols or manual configuration, which made scaling to millions of vehicles impractical. ISO 15118-20 was designed for interoperability from the ground up, meaning that any charger compliant with the standard should work with any vehicle that supports it, regardless of manufacturer. This is precisely why Hyundai's successful test is such a significant milestone: it proves that the standard works with a commercially available vehicle from a major global OEM, removing one of the largest remaining technical barriers to widespread V2G deployment.

The Bigger Picture for Energy and Transportation

Vehicle-to-Grid technology has been promised since the earliest days of the CHAdeMO DC charging standard, which Nissan and Mitsubishi specifically designed for bidirectional power flow. The original Nissan Leaf from 2010 was technically V2G capable, yet the technology never reached mass adoption because the supporting infrastructure consistently lagged behind the vehicles. CCS—the standard that replaced CHAdeMO in most markets—was slow to include bidirectional specifications, charger hardware was inconsistent and expensive, connection standards were disputed across regions, and, most importantly, vehicle manufacturers were reluctant to warranty their batteries for continuous discharge into the grid. Hyundai's successful test in Australia removes a major barrier by demonstrating manufacturer confidence that bidirectional discharge can be done safely without voiding vehicle warranties. The company noted that any vehicle built on its 800V Electric-Global Modular Platform could theoretically be upgraded for V2G through a software update, meaning the Ioniq 5 and Ioniq 6 may gain the feature pending regulatory and manufacturer approval. ARENA forecasts that up to 2.6 million Australian homes could adopt V2G by 2040, making the country one of the most ambitious markets for the technology and creating a proving ground that other nations can learn from.

6. Looking Ahead

These stories share a common rhythm that is becoming familiar in the technology industry: the gap between a technology's promise and its real-world deployment keeps narrowing, but not evenly and not predictably. AI governance is racing to catch up with models that outpace the rules, and the Anthropic-Mythos standoff showed how quickly government regulators can pivot from a hands-off approach to direct intervention when they perceive a national security risk, even when the technical substance of that risk is contested. EV autonomy is advancing faster in controlled, incremental features like parking memory and voice commands than in its ultimate goal of unsupervised operation, and Europe's regulatory scrutiny of Tesla's speed-limit behavior could create a harder ceiling on FSD claims than any engineering limitation. Battery innovation is finally moving from lab demos to factory partnerships, with Honda and QuantumScape providing a concrete timeline for solid-state commercialization that the industry has been waiting years to hear. Hyundai's V2G demonstration shows that bidirectional vehicle-grid integration is technically ready and waiting only for manufacturers, regulators, and energy companies to align on standards and adoption frameworks. Meanwhile, the AI talent movement—from Shazeer's move to OpenAI to the broader debate about geopolitical competitiveness—shows that the people building these systems are making strategic bets about where real progress will happen over the next decade. The next six months will reveal which promises become products and which stall at the demo stage again, and smart observers will watch for the signals that separate durable progress from marketing momentum.