AI News August 15, 2026 8 min read 6 sources

AI News August 15, 2026: Gemini 3.7 Flash Lands, DeepSeek V4 Pro Goes Official, Claude Code Flips to Auto Mode, GPT-5.6-Cyber Finds Chrome Zero-Days, Manus Unwinds Meta Deal

Google ships a faster Flash for agents at half price. DeepSeek's V4 Pro exits preview with new pricing. Anthropic turns autonomy on by default in Claude Code. OpenAI's cyber model digs up two Chrome zero-days. And Manus prepares to go independent again as its Meta acquisition collapses under a Beijing order.

⚡ Top 5 AI Stories — August 15, 2026

The AI news cycle is now measured in days, not weeks. This week alone: Google shipped a new Flash model three weeks after the last one, DeepSeek pushed its flagship out of preview, Anthropic redefined how developers interact with Claude Code, OpenAI let a cybersecurity-specialized model loose on Chrome — and it found real bugs. Meanwhile, Manus is set to become independent again after one of the strangest acquisition unwinds in tech history. Here’s what matters.


1. ⚡ Google Launches Gemini 3.7 Flash — Built for Coding and AI Agents

Google released Gemini 3.7 Flash on August 13, just three weeks after Gemini 3.6 Flash — an unusually fast turnaround the company attributes to developer feedback and algorithmic improvements.

Why it matters:

  • Coding-first design: Google DeepMind says 3.7 Flash shows major gains in debugging and issue resolution, generates more functional web layouts with fewer prompts, and improves reasoning on real-world business workflows
  • Half-price launch: Through the end of 2026, Gemini 3.7 Flash costs $0.75 per million input tokens and $3.75 per million output tokens — a 50% introductory cut
  • Agentic focus: The model is optimized for planning, tool use, and multi-step workflow execution, and is rolling out to Gemini Spark, Google’s agent service for AI Pro and Ultra subscribers in 160+ countries
  • The elephant in the room: Google still gave no date for its delayed flagship Gemini 3.5 Pro, even as CEO Sundar Pichai confirmed compute is already being spent training Gemini 4

The message is clear: while the flagship is stuck, Google is competing on iteration speed and price — a direct challenge to the assumption that frontier quality requires frontier prices.


2. 🐋 DeepSeek V4 Pro Goes Generally Available — With a Pricing Twist

On August 13, DeepSeek officially released DeepSeek-V4-Pro-0813, moving its flagship out of preview four months after the model family first appeared.

The GA build brings substantial agent upgrades:

  • Flexible reasoning effort: low for simple tasks, high for daily agent workflows, and max for complex problems — on both V4-Pro and V4-Flash
  • Native OpenAI Responses API support, optimized for Codex with one-click setup
  • A fast-improving curve: Reuters notes DeepSeek’s cheaper V4 Flash unexpectedly beat the April V4 Pro preview in several independent tests — and the official Pro now leapfrogs that

The most interesting move is economic: starting August 16 at 16:00 UTC, DeepSeek switches to peak/off-peak API pricing, with off-peak rates at 50% of peak — a scheduling incentive no Western frontier lab currently matches. Combined with the MIT-licensed open weights of V4 Flash, DeepSeek keeps applying downward pressure on the entire market’s pricing.


3. 🤖 Anthropic Flips Claude Code to Auto Mode by Default

Starting August 14, new Claude Code sessions on Pro, Max, and Team plans run in auto mode by default — replacing per-action approval prompts with a classifier that vets every tool call for irreversible, destructive, or out-of-bounds actions.

Anthropic’s pitch is blunt: the classifier is better at this than you are.

  • In internal testing across 1,000+ paid users, the classifier caught 89% of dangerous commands versus just 13.6% for human reviewers clicking through prompts
  • Teams using auto mode shipped roughly 25% more pull requests
  • The extra tokens the classifier burns are no longer charged to Pro, Max, and Team usage limits
  • Enterprise and API users stay opt-in “for now,” with a default flip planned within a month

It’s a milestone in the shift from human-in-the-loop to human-on-the-loop — and an acknowledgment that approval fatigue, not capability, was the real bottleneck in agentic coding.


4. 🔐 OpenAI’s GPT-5.6-Cyber Found Two Chrome Zero-Days

OpenAI’s Daybreak security initiative expanded on August 10 with a purpose-trained model: GPT-5.6-Cyber, which answers 95% of sensitive security queries covering exploit-chain development, authentication bypass, and privilege escalation — up from 57.3% for its predecessor.

Its first scalp is real: the model discovered two previously unknown V8 vulnerabilities in Chrome that can be chained to corrupt memory and bypass the V8 heap sandbox. Google patched them under CVE-2026-15903. GPT-5.6-Cyber is OpenAI’s first model to hit the “High” cyber capability threshold under its Preparedness Framework, and hardware security keys become mandatory for all Daybreak accounts on September 1.


5. 🪞 Manus Will Return as an Independent Company as Meta Deal Unwinds

Manus confirmed on August 11 that it will “soon return to operating as an independent company” as its Meta acquisition unwinds under a Beijing order.

  • Data generated by certain users on or after December 29, 2025 (Meta’s acquisition date) will be deleted between August 23–24, 2026 — backups must be completed by 7:59 a.m. SGT on August 23
  • Manus stresses the split is regulatory compliance, not a security incident
  • Unaffected users can continue using the service normally through the transition

A billion-dollar AI acquisition being forcibly reversed by regulators — with user data caught in the blast radius — is a preview of how geopolitics will increasingly shape who owns what in AI.


📊 The Day in Numbers

MetricValue
Gemini 3.7 Flash intro pricing$0.75 / $3.75 per million tokens
Gap between Gemini 3.6 and 3.7 Flash3 weeks
DeepSeek off-peak discount50% (effective Aug 16)
Claude Code classifier catch rate89% vs 13.6% human review
Chrome zero-days patchedCVE-2026-15903 (2 V8 flaws)
Manus user-data deletion windowAug 23–24, 2026

🔮 What to Watch Next

  • Gemini 3.5 Pro: Every Flash release without a Pro date raises the pressure on Google’s flagship roadmap
  • DeepSeek pricing ripple: Watch whether OpenAI or Anthropic respond to time-based pricing
  • Claude Code Enterprise flip: Anthropic says auto mode defaults reach Enterprise and API within a month
  • Daybreak expansion: More purpose-trained cyber models are almost certainly coming — from OpenAI and rivals alike
  • Manus independence: Whether it survives, rebrands, or gets re-acquired will test how durable forced unwinds are

This article was compiled from multiple sources on August 15, 2026. All figures and claims are attributed to the original publications linked in the sources list above.

#google#gemini-3-7-flash#deepseek#v4-pro#anthropic#claude-code#auto-mode#openai#gpt-5-6-cyber#chrome-zero-day#manus#meta#regulation#ai-agents#open-source