5 PM Technology News Recap: Revolut Data Exposure, AI Pacing, LG TV Privacy, GitHub Copilot, and StarCraft (September 12, 2026)

1. Revolut confirms customer data was exposed through fake government requests

Revolut confirmed that it disclosed sensitive customer information to an unauthorized third party after fraudulent information requests arrived from a legitimate government-agency email domain.^[1][2] Reuters and TechCrunch both attribute the confirmation to a Revolut spokesperson.

According to an affected-customer notice reviewed by TechCrunch, exposed data included birth dates, postal and email addresses, phone numbers, and identity-document copies such as passports and driver’s licenses. The notice said some records may also have included verification selfies, account statements, and transaction histories.^[1]

Revolut called the affected population “limited” but did not provide a number, identify the government agency, or say whether one market was affected. The company said it blocked the address, notified the relevant agency, enforcement bodies, data-protection authorities and financial regulators, and that its systems and customer funds were unaffected.^[1][2]

Practical action: affected customers should distrust messages that create urgency, verify contacts through Revolut’s official app or website, monitor financial and identity records, replace an exposed identity document when the issuing authority recommends it, and preserve the notice for dispute or fraud reporting. Never use a link or phone number supplied by an unexpected caller claiming to “help” with the breach.

2. Anthropic’s CEO proposes “pacing” frontier AI development

Anthropic CEO Dario Amodei published “We Must Pace the Frontier,” arguing that leading labs should slow the rate at which they improve model capabilities so safety work has more time to catch up. He explicitly says “pacing” does not mean halting model training or technical progress.^[3]

His three-part proposal calls for embedded third-party evaluators with substantial access inside frontier labs, government-enabled coordination among leading companies on safety standards and limits, and eventual coordination with geopolitical competitors on narrowly dangerous uses.^[3][4] Amodei says Anthropic is unilaterally committing to the embedded-evaluator step; TechCrunch reports OpenAI CEO Sam Altman said OpenAI would also do so, with details still to come.^[4]

The Associated Press reports that Amodei warned advanced AI could reach capabilities within six to 12 months that create severe cyber risks. That is his forecast, not an independently verified countdown or a consensus timeline.^[5]

Practical context: organizations buying AI systems do not need to wait for global agreements. They can demand evaluation summaries, incident-disclosure terms, tool and data boundaries, human approval for consequential actions, revocation controls, audit logs, and clear model-change notices in contracts now.

3. LG answers smart-TV spying allegations—but the dispute is not closed

LG says its televisions do not continuously record or transmit conversations. The company says far-field “Hi LG” listening must be enabled by the user, wake-word detection happens locally, and audio is deleted rather than sent to LG when the wake word is not recognized.^[6][7]

LG also says Automatic Content Recognition is opt-in and does not collect screenshots, screen recordings, video recordings, voice recordings, or other audio recordings. It confirmed that compatible-device discovery scans the local network, describing that behavior as a normal smart-TV function.^[6][7]

Those statements answer some allegations raised by Gamers Nexus and collaborating researchers, but the checked reporting says LG did not fully address broader data-sharing concerns, plain-text transcript claims, or all alleged vulnerabilities. Tom’s Hardware says it did not independently verify either the researchers’ allegations or LG’s counterclaims.^[6][7]

Practical action: review the television’s voice, ACR, advertising-personalization, viewing-data, and nearby-device options after every major update. Disable features you do not use, remove stale app permissions, apply firmware updates, and consider placing smart TVs on an isolated home network that cannot reach sensitive computers or storage devices.

4. GitHub gives enterprises more Copilot visibility while code review becomes more autonomous

GitHub says Copilot usage-metrics reports now include generally available fields for activity in the dedicated VS Code Agents window. Enterprise and organization reports can include daily active agent users, session counts, and total user messages; user-level reports can include whether a person used the window plus their session and message totals.^[8]

The company cautions that these fields cover the dedicated VS Code Agents window only, remain separate from editor-window Agent Mode and generic rollups, and may be absent or null when data is unavailable. Access is limited to specified owners, billing managers, or custom roles with the View Copilot Metrics permission.^[8]

In a near-window changelog published about 72 minutes before the prior recap cutoff, GitHub also said Copilot code review now resolves its own addressed comments, generates commit-message suggestions for applied fixes, uses more shell tools behind its agent firewall, and uses an ensemble of agents for Lite reviews.^[9] GitHub reports its own experiments increased addressed-comment counts for high-, medium-, and low-severity findings by 47%, 31%, and 11%, while reducing review cost by about 8%; those are company measurements, not an independent benchmark.

Practical action: grant metrics access narrowly; document what employee-level fields mean; keep branch protections and human review for risky changes; require logs for commands run by review agents; and sample automatically resolved threads to catch mistaken closure.

5. Blizzard reveals a new StarCraft with a 2030 release target

Blizzard’s official StarCraft site is live with an announcement cinematic titled “STARCRAFT | Announce Cinematic | Dominion” from the official StarCraft YouTube channel.^[10][11] The official page establishes the announcement and its Dominion framing; it does not publish a detailed release specification.

Independent event reports from The Verge and Shacknews say Blizzard announced the project at BlizzCon, described it as an open-world shooter rather than a traditional real-time strategy game, and targeted 2030.^[12][13] The Verge says the presentation showed a cinematic, not gameplay.

Practical context: fans can use Blizzard’s official page and channel to avoid impersonation and fake-preorder scams. Wait for actual gameplay, supported-platform details, accessibility information, business model, and firm release terms before spending money or planning hardware upgrades.