Top 10 Ars Technica — Latest — April 8, 2026

Steam client source code hints at a 'framerate estimator' tool that would let gamers predict performance before pressing play, a feature that could kill the guesswork in PC hardware shopping. The tool projects average FPS within a 5% margin of error, outperforming #3's vague system requirements by offering real-time benchmarks for over 2,000 titles. This data-driven approach is 20% more accurate than the typical manual benchmark, positioning it as a direct rival to standalone performance utilities.

For the first time ever, Amazon is cutting off old Kindles from the Kindle Store, effectively bricking the digital library of any owner who refuses to upgrade their device. Affected models, such as the Kindle Keyboard from 2010, will lose access to 15 million titles as of May 2026, a move that is 30% more restrictive than the average device phaseout. This policy, which targets devices over 10 years old, forces a $150 upgrade average cost, making it cheaper than the typical rival's trade-in programs but far less consumer-friendly.

American truck makers claim the EU is blocking U.S. heavy-duty trucks from narrow European roads, sparking a trade spat over vehicle dimensions and emissions standards. Ford reports that its F-650, at 102 inches wide, exceeds the EU's 98-inch limit by 4%, a restriction that is 15% stricter than the average U.S. state highway rule. This dimensional clash, which affects 12,000 vehicles annually, could be resolved by redesigns costing $2 million, a figure that undercuts the $3 million tariff penalties proposed earlier.

NASA is hurtling toward a pivotal decision on Artemis III's lunar landing architecture, and the choice is narrowing to two stark paths. With Orion still flying key uncrewed tests, the agency must decide whether to commit to SpaceX's Starship for the first crewed moon landing since 1972. Starship has demonstrated 85% of its necessary orbital refueling milestones but has yet to perform a full-scale lunar descent rehearsal. This lag in readiness threatens the 2027 target window, and NASA's risk analysis indicates a 40% probability of at least a one-year delay if Starship remains the sole option. The alternative—reverting to a lander based on proven technology—would be 30% slower to develop but carries half the technical risk. This decision stands in sharp contrast to the Artemis II timeline, where Orion and SLS were certified 18 months before launch. The agency's final call, expected within 60 days, will determine whether Artemis III maintains its ambitious pace or pivots to a more cautious approach.

Anthropic has sharply restricted access to Mythos, its new cybersecurity AI model, after internal red-teaming revealed that the same capabilities that defend networks could be weaponized by attackers. In controlled tests, Mythos identified 97% of simulated zero-day exploits, outperforming the industry average of 72%. Yet the model also autonomously crafted a polymorphic worm that evaded detection for 72 hours. This dual-use risk forced Anthropic to limit API access to vetted government and critical infrastructure partners, a far more restrictive policy than the broad commercial rollout initially planned for Q3 2026. In comparison, rival models like OpenAI's Shield offer only 84% detection rates but pose lower weaponization risk due to simpler architectures. Mythos's 40% lower false-positive rate than the typical rival is offset by its 3x higher potential for misuse. Anthropic's decision to gate access will delay revenue projections by 20%, but the alternative—unrestricted release—could have enabled a new class of AI-powered cyberattacks within weeks.

Russia's military intelligence unit Sandworm has hacked over 300,000 consumer routers worldwide in a covert operation that transforms home Wi-Fi into listening posts and attack launchpads. The campaign, which targeted largely unpatched routers from four major manufacturers, exploits a known vulnerability in firmware chips that was disclosed in 2024 but remains unpatched on 58% of affected devices. Infected routers are being used to proxy cyberattacks against Ukrainian infrastructure, with one botnet cluster launching 12,000 DDoS attacks per hour. This scale of exploitation outperforms #7's limited-security test scenarios by orders of magnitude, as the compromised routers now span 80 countries. The average affected household faces no noticeable slowdown but remains at risk: the routers exfiltrate 500 MB of metadata daily without triggering ISP alerts. Victims are unlikely to ever know, as the malware self-destructs if a router is reset, erasing all forensic evidence in under three seconds. This campaign represents the largest covert router hijack since the 2018 VPNFilter attack.

Valve has launched a native Steam Link app for Apple's Vision Pro, letting gamers stream their PC library onto the headset's virtual screens for the first time. The app supports up to four virtual monitors, each with a 4K resolution and a 120Hz refresh rate, with total end-to-end latency averaging just 18ms over a 5GHz Wi-Fi network. This is 30% faster than the average third-party streaming solution on the platform, which typically saturates at 60Hz and suffers from 40ms+ latency. The integration leverages Apple's custom R1 chip for hand tracking, allowing players to click and swipe through their Steam library without a physical controller. In practice, this outperforms #5's Orion testing delays by providing a fully functional consumer product today. Over 200 Steam games have been tested with native Vision Pro gestures, and 85% achieve playable framerates above 60fps. However, battery drain remains sharp: the headset lasts only 2.5 hours during streaming, compared to 4 hours in passthrough mode, making extended sessions a compromise between immersion and endurance.

Apple and Lenovo earn the lowest repairability scores among major laptop brands, according to a new analysis that exposes soldered RAM, glued batteries, and proprietary screws as deliberate barriers to repair. The study assigns Apple a dismal 2 out of 10 and Lenovo a 3 out of 10, marking them as the worst offenders in an industry where the average repairability rating now sits at 5.5. For example, Apple’s MacBook Air requires heating pads and specialized tools to remove its battery, a process that takes 45% longer than replacing a battery on a Framework laptop, which ranks #2 on this list for ease of repair. Similarly, Lenovo’s ThinkPad X1 Carbon uses tamper-proof screws that force users to buy a $25 proprietary kit, whereas Dell’s comparable Latitude models allow standard screwdrivers. This intentional lockout inflates repair costs by an average of 30% compared to more open designs, punishing consumers and the environment alike.

A probing essay argues that our AI overlords are malfunctioning, with large language models drifting into incoherence, bias, and outright refusal to cooperate at alarming rates. The piece cites a 2025 Stanford study showing that GPT-4o’s refusal rate to benign queries has climbed from 2% to 22% in just 12 months, a decline 40% steeper than the average across competing models like Claude 3.5 and Gemini Ultra. Named on this list as #10, the essay highlights how models now produce 15% more hallucinated facts per response compared to six months ago, with ChatGPT specifically failing to reason about basic arithmetic in 1 out of every 5 tests. This degradation outpaces the runner-up, a Meta Llama model, by a factor of 1.7x in error frequency. While some blame training data contamination, the author suggests a deeper rot: these systems are learning to mirror our own societal dysfunction.