June 27, 2026
Today's top tech stories, deduped across the newsletters I read and briefly summarized. Click a source to open the original article.
AI
OpenAI previews GPT-5.6 Sol — but the U.S. government decides who gets access
OpenAI has begun a limited preview of GPT-5.6 Sol (flagship), Terra (balanced), and Luna (fast/cheap), but access to Sol requires U.S. government vetting. Terra delivers GPT-5.5-level performance at 2× lower cost.
U.S. clears Anthropic to release Mythos model to trusted American organizations
The U.S. government has authorized Anthropic to give select trusted U.S. organizations access to its powerful Mythos AI model, in a controlled rollout that mirrors the government-vetted access framework now emerging for frontier AI.
NVIDIA quantizes Nemotron 3 Ultra to NVFP4 with Model Optimizer
NVIDIA has released an NVFP4 checkpoint for Nemotron 3 Ultra via NVIDIA Model Optimizer, compressing model weights to accelerate inference on long-context workloads with minimal accuracy loss.
Security
FBI: Russian hackers are now stealing Signal Backup Recovery Keys
The FBI and CISA warn that a Russian intelligence-linked phishing campaign has evolved beyond credential theft to stealing Signal Backup Recovery Keys, giving attackers full access to victims' message history.
Supply-chain attack costs Polymarket customers $3 million
Hackers injected a malicious script into Polymarket's frontend through a breach at a third-party vendor, resulting in approximately $3 million in customer losses. Polymarket says it will fully reimburse affected users.
CISA sets Sunday deadline to patch actively exploited Cisco vulnerability
CISA has ordered federal agencies to patch an actively exploited vulnerability in Cisco Unified Communications Manager Server by this Sunday, citing ongoing attacks in the wild.
Threat actors use fake OpenAI organizations to phish employees
Attackers are creating fraudulent OpenAI tenants impersonating legitimate companies and inviting employees to join, apparently to trick targets into sharing sensitive company information through AI chats and projects.
2,000 people tried to hack an AI assistant — here's what it revealed about prompt injection
Fernando Irarrázaval ran a public challenge inviting 2,000 participants to extract secrets from his AI legal assistant; the experiment exposed multiple prompt injection and jailbreak techniques that bypassed the system's defenses.
Dev
AWS Lambda introduces MicroVMs with full lifecycle control
AWS Lambda now supports MicroVMs, giving developers isolated sandboxes with start, pause, resume, and snapshot controls for workloads that require stronger isolation than standard Lambda functions.
Anthropic raises Claude API rate limits and simplifies usage tiers
Anthropic has raised rate limits across the Claude API so that Sonnet and Haiku now match Opus limits at every tier. Usage tiers are consolidated into three — Start, Build, and Scale — and no organization receives lower limits than before.
IT
EFF pushes back against California's 3D printer surveillance bill
California is advancing legislation that would require 3D printers to embed hidden tracking codes in all printed objects, raising serious privacy concerns. The EFF is urging the public to oppose the bill before it advances further.
PlayStation is deleting 551 movies from customers' digital libraries
Sony is removing 551 titles from PlayStation users' digital libraries after losing its distribution deal with Studio Canal, reigniting the debate over how fragile digital ownership of purchased content really is.