5 PM Technology News Recap: GitHub SHA-1, F5 Bot Defense, Google Language AI, ATLAS, and Apple Updates (September 15, 2026)

1. GitHub switches off SHA-1 in HTTPS for its cloud service

GitHub says it disabled SHA-1 in HTTPS for github.com and partner content-delivery networks on September 15, following the schedule announced in April. The change covers GitHub Enterprise Cloud, including Data Residency, while GitHub Enterprise Server is unaffected.^[1]

This retirement concerns SHA-1 use in HTTPS and TLS negotiation—not the familiar SHA-1-looking identifiers that still label many Git objects. GitHub’s earlier notice warned that older browsers, API clients, and Git clients using HTTPS could fail if they cannot negotiate newer algorithms.^[2]

Practical action: if a previously working clone, fetch, API call, or browser session now fails over HTTPS, record the exact TLS error, client version, operating system, proxy path, and certificate chain. Update the client or runtime through a supported channel before changing repository settings. Test the same endpoint from a known-modern client to separate a legacy-TLS problem from an account or network problem.

2. F5 adds agent-aware bot controls and device-risk scoring

F5 announced new agentic-AI protections for Distributed Cloud Bot Defense on September 15. F5 says the controls classify traffic from people, trusted AI agents, and malicious bots inside one policy framework, with actions such as allow, step-up challenge, rate-limit, or block.^[3]

The release also describes persistent device identification across sessions and accounts, plus real-time risk scoring for signals such as emulators, spoofing, and tampering. However, availability is split: F5 says the new agentic protections are available now, while device intelligence is expected to enter limited availability in the fourth quarter of 2026 before a broader rollout.^[3]

Help Net Security reported the same product package, but its account tracks F5’s announcement rather than publishing an independent detection-rate or false-positive test.^[4]

Practical action: teams evaluating bot defenses should replay known legitimate and abusive workflows, measure challenges and blocks by device and account, define an appeal or recovery path, and minimize retained device data. Keep payment, login, checkout, account-recovery, and API policies separately observable so one aggressive rule cannot silently damage every customer journey.

3. Google targets more useful AI across underrepresented languages

Google says its long-term goal is to support the world’s 1,000 most-spoken languages and describes moving beyond literal text translation toward systems that handle speech, tone, slang, and local expression. Google says its Universal Speech Model was trained on 12 million hours of audio and uses cross-lingual transfer learning to improve recognition where training data is limited.^[5]

The company also describes community partnerships for open language data and emphasizes offline or low-connectivity use. Those design choices matter because the web overrepresents a small group of dominant languages, but the September 15 article is a company research summary—not a complete independent accuracy audit for every language, accent, dialect, and context.^[5]

Practical action: organizations deploying translation or speech AI should test with native speakers from the actual service population, publish a manual fallback, preserve the original message, and measure errors separately by language and task. High-stakes medical, legal, financial, and emergency communications still need qualified human review.

4. Google opens a more explorable AI-and-work data atlas

Google launched a new interactive, open-access experience for its AI & Economy ATLAS on September 15. The company says the interface exposes millions of global data points about AI use by occupation, country, and personal context.^[6]

Google highlights several findings from its data and related work with Google DeepMind and MIT FutureTech: computer and mathematical occupations account for 30% of work-related AI use in the United States, and surveyed scientists report saving almost seven hours per week with AI. Google also says nearly half of surveyed scientists use some form of AI every day.^[6]

Those figures require denominator discipline. They describe Google’s dataset and cited survey, not every worker or scientist. Self-reported time savings are not the same as measured productivity, quality, safety, publication speed, or economic gain.

Practical action: managers can use the atlas to choose where to investigate, then run a local baseline: task completion time, rework, error severity, review load, security exceptions, and user outcomes. Keep “hours saved” separate from “value created,” because faster low-quality work can increase downstream cost.

5. Apple’s new platform releases arrive—with feature and usage limits

Apple’s September 14 update says major releases for iOS 27, iPadOS 27, macOS 27, and other platforms began rolling out, including Siri AI, new Apple Intelligence features, parental controls, and platform performance changes.^[7]

Apple reports that, in its testing, apps on iPhone and iPad can launch up to 30% faster, photos can load up to 70% faster after capture, and AirDrop transfers can be up to 80% faster. Those are vendor measurements with “up to” language, not guaranteed gains for every supported device, network, app, or workload.^[7]

The fine print matters: Apple says some server-side Apple Intelligence features are subject to daily usage limits that can vary by feature, request complexity, demand, and policy. Apple also says expanded access will be offered for a fee in the future, and Siri AI is not available to users under 13.^[7]

Practical action: before a broad deployment, confirm device eligibility, available storage, backups, line-of-business app support, authentication flows, VPN behavior, and management policy. Pilot the update with a reversible group, and tell users that cloud AI availability and usage limits may differ from local features.