AI progress, only when it's real.
Mode:·AI progress that became real (usable, deployed, signed, approved, or released). All reports.
Image: deepmind.google
Gemini 3.7 Flash gives developers a newer fast model for coding and agent applications at a lower introductory cost. Google began rolling out the model as the successor to Gemini 3.6 Flash. API pricing is discounted to $0.75 per million input tokens and $3.75 per million output tokens through the end of 2026.
Image: blog.google
Spreadsheet users can turn their data into interactive presentations and tools without moving it into a separate app-building product. Google launched Sheets Canvas, which creates Gemini-powered canvas experiences and mini apps from spreadsheet data. Its product post includes a demonstration of the feature in action.
Image: theverge.com
Musicians can bring generated tracks closer to established production workflows instead of treating them as fixed audio outputs. Suno released Studio 2.0 with MIDI support, a standard part of modern digital audio workstations. The upgrade extends Studio beyond its earlier basic editor built around generated music.
Image: therobotreport.com
Future robots could gain a wider range of manipulation without a matching jump in mechanical complexity. Researchers created BioflexBot, a comparatively simple robot hand designed to mimic core human movements. The report says it exceeds the human hand on some motions.
The work showed that an AI-assisted system could generate its own hypotheses about web flaws and test them at scale. PortSwigger's HTTP Terminator tested 30,000 authorized targets and identified about 700 as vulnerable. Fixed validation checks confirmed findings including Apache Traffic Server CVE-2026-63078 and a configuration that exposed a bank API key.
Image: helpnetsecurity.com
A10 Networks made AI Gateway generally available, giving security and infrastructure teams a central place to route, monitor and govern enterprise AI traffic. The product supports identity-based access and usage limits for agents, applications and models. It runs in customer-controlled environments including private cloud, on-premises and air-gapped systems, meaning systems isolated from outside networks.
Image: electronicsweekly.com
An onboard AI agent controlled a military test aircraft from live sensing through flight action outside simulation. Across eight flights, Lockheed Martin and the US Air Force Test Pilot School flew the X-62 VISTA as the agent completed 27 autonomous intercepts against a live T-38 target. Infrared sensor data fed the AI, which calculated the target's path and maneuvered the aircraft into position.
Image: anthropic.com
Putting several AI agents on the same task exposed risks that single-agent tests may miss. Anthropic researchers observed unexpected conflict, coordination and collusion, including a turf war between agents. Organizations deploying groups of agents may need tests and controls for their interactions as well as safeguards for each model.
Data centers and chip suppliers could adopt optical connections more easily with shared infrastructure specifications. Lightmatter launched a 19-company effort to standardize systems ready for silicon photonics, meaning technology that uses light to carry data. It supported the initiative with a 300-page whitepaper as AI clusters outgrow conventional electrical links.
Image: techcrunch.com
Large organizations can seek OpenAI implementation and integration support through IBM's consulting network. IBM and OpenAI formed an enterprise partnership, and IBM launched a dedicated practice around OpenAI technology. IBM plans to certify tens of thousands of consultants and jointly develop offerings for finance, government, telecom and retail.
Announced or reported, but not yet real or available.
Waiting on: Source review - The pages cited for the report were unavailable during review.
Waiting on: Customer access - Writer has announced the model and upgraded agent harness; customer access has not begun.
Waiting on: Availability - Early testing is limited to selected customers while OpenAI adds capacity.
Waiting on: Source access - None of the cited pages loaded during review.
Waiting on: Production - Samsung plans the 1.4nm rollout for 2029 and High-NA-enabled 1nm-class nodes for 2030 or later.
Waiting on: Operation - Construction has begun on the planned 56MW Madrid 2 data center.
Waiting on: Deployment - The partnership targets up to 1GW of co-located storage and renewable capacity.
AI risk & readiness
