The latest AI news from Anthropic, OpenAI, AWS, Google, Microsoft, Nvidia, Mistral, Recursive, Hugging Face, and the US Administration, and an AI customer reference from Shopify.
VOL 3 AUGUST 2026
INFO-TECH AI INSIGHTS
AI Transformation Brief
Featuring AI best practices and insights to enable our members to strategize, plan, develop, deploy, manage, and govern AI-based technologies and solutions.
In This Issue AI in the News | AI Research Highlights | Vendor Spotlight | Upcoming Events & Resources
AI IN THE NEWS
Anthropic to codesign AI chips, names their first Chief Global Affairs Officer, and discloses Claude’s security breach
Claude’s unauthorized access disclosure
On August 5, 2026, Anthropic confirmed it will codesign its own AI chips.
On August 4, 2026, Anthropic named Tino Cuéllar its first Chief Global Affairs Officer.
On July 30, 2026, Anthropic disclosed that Claude had gained unauthorized access to three organizations’ systems during cybersecurity evaluations.
ANALYST ANALYSIS
Taking a page from Google’s and AWS’s playbook, Anthropic’s initiative to build their own chips demonstrates they want to reduce their dependency on Nvidia and Google TPUs for compute infrastructure.
Creating the role of a Chief Global Affairs Officer reflects how critical government relations have become. This also signals that Anthropic is likely preparing for more regulatory battles as AI regulations continue to be introduced.
The disclosure that three Claude models breached real organizations during cybersecurity tests mirrors OpenAI’s own incident. Unlike OpenAI’s incident, no human disabled the guardrails here; the models triggered the breach on their own during a routine evaluation.
OpenAI joins AI safety evaluations, unveils Astra, and discloses Hugging Face breach
On August 3, 2026, OpenAI joined the White House’s voluntary AI safety-testing framework alongside Anthropic and Google.
On August 1, 2026, OpenAI revealed Astra, its next major model, after an internal version solved ten decades-old math problems.
On July 21, 2026, OpenAI disclosed that its own model breached Hugging Face’s systems during a cybersecurity evaluation.
ANALYST ANALYSIS
While the frontier model providers have been requested to provide input regarding AI model safety, the focus at this time appears to be on cybersecurity. AI safety is much more than just cybersecurity, and the lack of transparency in the administration’s process is unlikely to improve safety initiatives.
OpenAI unveiled Astra via the announcement that the model produced ten decades-old math proofs, at the cost of only $2,000 in compute. In addition, after applying their internal Preparedness Framework, OpenAI acknowledged that Astra demonstrated “Critical” cybersecurity capabilities.
OpenAI's own model breaking out of a guardrails-off test to autonomously hack Hugging Face and steal the benchmark answers is the industry’s first documented loss-of-control incident, undercutting its safety credibility with regulators.
#2
INFO-TECH AI INSIGHTS
AI Transformation Brief
AI IN THE NEWS
AWS launches AgentCore Harness, ships AgentCore Observability, and Recursive signs $410 million, multi-year collaboration with AWS to scale self-improving AI
Autonomous Business Insights Framework
On August 5, 2026, AWS announced general availability of Amazon Bedrock AgentCore Harness.
On July 31, 2026, AWS detailed new Amazon Bedrock AgentCore Observability tools for production agents.
On July 28, 2026, Recursive signs a $410 million deal to scale self-improving AI with AWS.
ANALYST ANALYSIS
AgentCore Harness collapses agent orchestration into two API calls, undercutting the do-it-yourself stitching of Bedrock, Lambda, and open-source frameworks that AWS customers relied on before. It also folds orchestration in-house, sidelining third-party agent frameworks.
Deeper Observability tooling shows AWS betting that production debugging, not just model quality, is what decides whether enterprises keep agents running at scale. This addresses of the top requirements that enterprises cite for running agents in production.
Recursive’s $410M multi-year AWS deal shows cloud providers underwriting compute for pre-product research labs, betting the payoff from hosting a future frontier lab outweighs the risk that self-improving AI falters before shipping. It’s also a proof point for AWS’s broader AI-investment strategy.
Google ships Gemini Robotics 2, the Science One Framework, and new Gemini 3.6 Flash models
Gemini 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber
On July 30, 2026, Google DeepMind announces Gemini Robotics 2 brings whole body intelligence to robots.
On July 30, 2026, Google Research unveils their “Science One Framework” for Verifiable AI Research.
On July 21, 2026, Google announced Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber.
ANALYST ANALYSIS
Gemini Robotics 2’s full-body control, demonstrated on Apptronik’s Apollo 2, propels DeepMind beyond tabletop manipulation and into the humanoid market already currently dominated by Nvidia.
The Science One Framework’s push for verifiable, chain-of-evidence AI research positions Gemini as a scientific-discovery tool, not just a chatbot, ahead of rivals’ own science claims.
Splitting the Flash line into a gated 3.5 Flash Cyber for vulnerability hunting mirrors the industry’s shift toward specialized, access-controlled security models alongside general-purpose assistants.
Microsoft releases Orchard, doubles down on a Copilot super app, and expands its Mistral AI partnership
On August 4, 2026, Microsoft released Orchard, an open-source framework for agentic AI.
On July 29, 2026, Microsoft doubled down on multi-model AI as it builds a Copilot super app.
On July 21, 2026, Microsoft expanded its European compute and sovereign AI deal with Mistral AI.
ANALYST ANALYSIS
Open-sourcing Orchard lets Microsoft Research claim state-of-the-art agent training results while ceding some infrastructure advantage to the community, a bet that ecosystem goodwill outweighs the head start it gives rivals. It also positions Microsoft as an agent-training standard.
A Copilot super app consolidating chat, Cowork, and Autopilots mirrors OpenAI’s and Anthropic’s own consolidation moves, signaling that fragmented AI products are becoming a liability rather than a feature. It also pressures OpenAI and Google to ship unified assistants.
Deepening the Mistral partnership gives Microsoft a sovereign-AI solution in Europe without building its own frontier model there, hedging against regulatory pressure that increasingly favors non-US compute and models.
#3
INFO-TECH AI INSIGHTS
AI Transformation Brief
AI IN THE NEWS
White House delivers its AI oversight framework and will exempt “open” AI systems from security review
Read the Washington Post exempting open AI systems article
On August 3, 2026, the White House communicated that it had met its deadline to establish a voluntary framework for evaluating advanced AI models.
On August 4, 2026, the White House informed top AI companies it will exempt open-weight AI models from its planned prerelease security review, focusing scrutiny on closed, proprietary frontier systems instead.
ANALYST ANALYSIS
On June 2, 2026, an executive order to create an AI oversight framework established an August 1 deadline for federal agencies to create a prerelease evaluation framework for advanced "frontier" AI models. By late July, the major AI labs (including OpenAI, Google, and Anthropic) were requested to review draft versions of the framework and submit feedback/edits, and after administration claimed that the framework was delivered on time, meeting the August 1 deadline.
However, there has been no communication or disclosure what the framework contains, who has seen it, nor when companies can start using it. Policy analysts argue that combining secret evaluation metrics with strict deployment controls grants the government an administrative “kill switch” over the AI models. By delaying or restricting market entry behind closed doors without external verification, the administration holds total discretion to suppress or block its AI systems before public release.
Whatever the reason behind these actions and the lack of transparency, this has only served to fuel the drive for sovereign AI systems around the world to reduce dependency and risk on AI models that the administration has jurisdiction over.
Nvidia announces $500 billion financing partnerships to drive AI compute infrastructure
On August 10, 2026, Nvidia established an AI compute infrastructure financing partnership to mobilize over $500 billion of third-party capital.
ANALYST ANALYSIS
Nvidia signed MOUs with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to build compute financing platforms, mobilizing over $500 billion of third-party capital, turning Nvidia GPUs into an investable asset class rather than a line item on CapEx budgets.
This $500B model funds AI clusters as utility infrastructure using institutional private capital, rather than venture equity or budgets from public institutions.
By coupling GPU hardware directly to private credit platforms, Nvidia creates a financial moat that makes it nearly impossible for competitors (such as AMD or custom hyperscaler ASICs) to win orders, regardless of chip pricing.
Shopify’s AI tools triple traffic, lift Q2 profits and revenue
On August 5, 2026, Shopify reported Q2 revenue up 36% to $3.6 billion, crediting AI-driven traffic and orders, which tripled year-over-year, as a key driver.
ANALYST ANALYSIS
Shopify’s framing cuts against the narrative hurting publishers: AI agents are complementing search, not cannibalizing it – traditional search sessions are still up 1.3x over two years.
Agents query Shopify’s catalog across many attributes instead of ranking by keyword, which is why 75% of AI-attributed purchases fall outside its top 100 “sweet spot” categories.
Building connectors into Claude, ChatGPT, Perplexity, and Vercel rather than competing with them positions Shopify as the checkout layer for agentic commerce regardless of which AI assistant wins.
#4
INFO-TECH AI INSIGHTS
AI Transformation Brief
AI RESEARCH HIGHLIGHTS
RESEARCH SPOTLIGHT:
Defend Against Defects and Technical Debt in Your AI-Generated Code
New!
Source: Defend Against Defects and Technical Debt in Your AI-Generated Code
Info-Tech Research Group
VENDOR SPOTLIGHT:
- Esri’s Customers Prove CIOs Are Getting Half the Story Without GIS
- Skyhigh Security SSE Platform: Hybrid Enforcement, Browser-Native AI Controls, and Integrated Data Security Posture
- AWS Seeks to Remove the Friction Around Government AI Adoption
- ThreatLocker: Default-Deny Endpoint Protection for the Post-Mythos Threat Environment
- Varonis Atlas AI Security Platform – End-to-End AI Security and Governance Platform
UPCOMING EVENTS
Info-Tech LIVE Event Registration
LIVE Barcelona - Barcelona, September 22-23, 2026
LIVE Toronto - Las Vegas, November 10-12, 2026
LIVE New Orleans - New Orleans, January 26-28, 2026
LIVE Gold Coast - Brisbane, March 16-17, 2027
LIVE Las Vegas - Las Vegas, May 11-13, 2027
Info-Tech IGNITE Event Registration
IGNITE Austin - Austin, September 17-18, 2026
IGNITE Washington DC – Washinton DC, October 1-2, 2026
IGNITE Sacramento – Sacramento, October 27-28, 2026
IGNITE Atlanta - Atlanta, November 2-3, 2027
IGNITE Vancouver – Vancouver BC, March 31-April 1, 2027
AI AND DATA ANALYTICS RESOURCES
Artificial Intelligence Research Center
AI Workforce Development Program
AI Workshops
AI Strategy
AI Governance, Liability & Risk
Accelerate adoption of Gen AI
Drive Business Value With M365 Copilot
Design Your Agentic AI Prototype
Develop Your Agentic AI Prototype
Optimize & Scale Your Agentic AI Prototype
AI EDITOR-IN-CHIEF
Bill Wong – Info-Tech AI Research Fellow
