Sinyalu MP bankruptcy, Hurupay exits Kenya, World Bank delays loan

Independent benchmarks from Inscribe and Dictato show Apple's SpeechAnalyzer, introduced at WWDC 2025, matching or beating OpenAI Whisper on accuracy for clean speech in French, Spanish, German, and Italian, while running entirely on-device with sub-300ms cold start latency. SpeechAnalyzer replaces the aging SFSpeechRecognizer on iOS and macOS and costs developers nothing in API fees or cloud round-trips. For anyone building speech features, Apple now offers a compelling on-device alternative to cloud Whisper and NVIDIA Parakeet, with the added benefit of privacy-preserving local processing.

Meta's ads engineering team published how they used sched_ext, the upstream BPF-based extensible scheduling framework, to build a custom scheduling policy for their ad serving fleet. When a routine Linux kernel upgrade threatened latency regressions across Meta's ad infrastructure, they wrote a custom BPF scheduler using sched_ext instead of reverting the kernel or patching around the problem. The approach shows that sched_ext, merged upstream in Linux 6.12, enables production teams to tailor scheduling behavior without kernel rebuilds. For any shop running latency-sensitive workloads at hyperscale, this is a reference architecture worth studying.

The Los Angeles Police Department has ended its data-sharing agreement with Flock Safety, a surveillance company that operates thousands of automated license plate recognition cameras across the US. The contract expired July 11 and was not renewed after the LAPD cited serious concerns over civil liberties and privacy. Flock had faced growing backlash over its network functioning as a citywide vehicle tracking dragnet. The decision is a rare reversal of a major police surveillance program on privacy grounds and puts pressure on other departments using similar technology across the country.

CISA, the FBI, and international partners issued a joint cybersecurity advisory warning that Russian Federal Security Service (FSB) Center 16 cyber actors are actively exploiting poorly configured and vulnerable networking devices worldwide, targeting critical infrastructure sectors. The advisory outlines five priority actions: implement SNMPv3, use strong unique passwords in place of defaults, disable legacy protocols like Cisco Smart Install, segment networks, and keep firmware updated. Any organization running older Cisco or legacy networking gear should treat this as an operational priority. The advisory builds on years of similar warnings but reflects an escalation in observed targeting.

Xinzhou Wu, Nvidia's vice president and head of automotive, revealed on The Verge's podcast that even his own division has to fight weekly battles for GPU compute allocation inside the company. If the head of automotive at the world's largest AI chip company cannot secure guaranteed compute for his own team, every other industry is feeling the pinch even harder. It is a rare window into the internal allocation wars that define the AI hardware economy in 2026, where even the GPU maker itself rations access.