The AI industry is showing signs of maturation across multiple fronts, from air cargo logistics to model benchmarking and consumer-facing regulation. At the Air Cargo Forum in Miami, Boeing's 2026 World Air Cargo Forecast predicts that AI hardware and advanced manufacturing shipments will double air cargo traffic over the next 20 years, with the global freighter fleet growing from 2,395 planes in 2025 to over 4,000 by 2045. However, Rotate CEO Ryan Keyrouse noted that while cloud computing exports from Asia Pacific tripled in 2025, future growth expectations have become more moderate.
In AI model evaluation, Artificial Analysis released its Intelligence Index v5, the largest update since its launch. Claude Opus 5.5 leads the ranking at 58 points, followed by GPT-6 Astra at 57 and GPT-6.1 Sol at 55. The gap between open weights models and top proprietary models has widened significantly from 3 points to 12 points. Two new benchmarks were added, including FrontierCode 2.0 for coding tasks and Terminal-Bench-Science for scientific research. Running costs for top models vary by more than 13 times per task.
Tesla has quietly renamed its Full Self-Driving computers across owner manuals for the Model 3, Model Y, and Cybertruck. The FSD Computer label now appears as SI Computer, with HW2.0, HW2.5, HW3, and AI4 becoming SI2.0, SI2.5, SI3, and SI4 respectively. This change follows Elon Musk's broader shift from the term AI to Super Intelligence. Although not officially announced, the updated manuals suggest the new labels may appear on vehicle screens through a future software update.
California's AB 1609 takes effect January 1, 2027, requiring businesses with over $500 million in revenue to connect a customer with a person within 15 minutes when using AI customer service. Fines can reach $5,000 per violation. Salesforce, Zendesk, and Genesys are already building the human handoff feature into their AI service products. New York City has also proposed a rule requiring businesses to keep records of automated interactions for three years.
At the Dubai AI Festival, over 20,000 participants from 100 countries gathered with 500 investors and 200 exhibitors. More than 100 Chinese technology companies and 23 Indian firms showcased their work. The Dubai International Finance Center AI campus has attracted more than 500 AI companies and over AED 1 billion in investment since May 2024. Scale AI opened a new office at DIFC and signed a partnership with Abu Dhabi-based Presight to develop AI solutions for government and enterprise customers.
Sonar released a Hunter Agent that hunts for logic bugs in code that traditional SAST tools often miss. In a demo scanning a Python app in about one hour, it found 46 open issues including a high severity IDOR vulnerability. Scheduled scans can run daily, weekly, or monthly, and automatic remediation through a SonarQube agent becomes available this week. Separately, researchers launched R2A-Sales, a benchmark revealing that AI sales agents follow rules in theory but fail in practice, with static compliance scoring 89.9 to 99.1 percent while grounded action compliance fell to only 3.0 to 17.3 percent.
Key Takeaways
- Boeing predicts AI hardware shipments will double air cargo traffic over 20 years, with freighter fleet growing from 2,395 planes to over 4,000 by 2045
- Claude Opus 5.5 leads Artificial Analysis Intelligence Index v5 at 58 points, ahead of GPT-6 Astra at 57 and GPT-6.1 Sol at 55
- Tesla renames FSD computers to SI Computer across Model 3, Model Y, and Cybertruck manuals, reflecting Musk's shift from AI to Super Intelligence
- California AB 1609 requires businesses with over $500 million revenue to hand off to a human within 15 minutes on AI customer service starting January 1, 2027
- Scale AI opened a new office at Dubai's DIFC and partnered with Presight to develop AI solutions for government and enterprise customers
- Sonar's Hunter Agent found 46 open issues including a high severity IDOR vulnerability by scanning a Python app in about one hour
- R2A-Sales benchmark shows AI sales agents have a compliance gap of 80.4 to 92.5 percentage points between knowing rules and following them
- Artificial Analysis Intelligence Index v5 shows running costs for top models vary by more than 13 times per task
- Dubai AI Festival attracted 20,000 participants from 100 countries with over 500 AI companies at the DIFC campus
- BehaveRec framework finds that long-term behavioral history helps recommendations after about 20 interactions per user
AI air cargo growth slows as demand moderates
Demand for AI-related air cargo shipments is slowing down despite continued investment in the sector. Ryan Keyrouse, CEO of Rotate, spoke at the Air Cargo Forum in Miami about the trend. He noted that cloud computing exports from Asia Pacific tripled in 2025. However, future growth expectations have become more moderate.
Boeing forecast says AI investment will boost air cargo
Boeing released its 2026 World Air Cargo Forecast at the Air Cargo Forum in Miami. The company predicts that AI hardware and advanced manufacturing shipments will double air cargo traffic over the next 20 years. Global air cargo traffic hit a record in 2025 with nearly 5% growth. Boeing expects the global freighter fleet to grow from 2,395 planes in 2025 to over 4,000 by 2045.
Artificial Analysis launches Intelligence Index v5
Artificial Analysis released version 5 of its Intelligence Index, a benchmark for frontier AI models. This is the biggest update since the index launched, with 60% of evaluations now using private datasets. Claude Opus 5.5 leads the ranking at 58 points, followed by GPT-6 Astra at 57 and GPT-6.1 Sol at 55. The gap between open weights models and top proprietary models has widened significantly from 3 points to 12 points.
Artificial Analysis announces Intelligence Index v5 details
The Intelligence Index v5 adds two new benchmarks and upgrades three existing ones. New additions include FrontierCode 2.0 for coding tasks and Terminal-Bench-Science for scientific research. Upgrades were made to HLE-Diamond, AA-Omniscience, and CritPt. Claude Opus 5.5 leads across most categories, with GPT-6 Astra close behind. Running costs for top models vary by more than 13 times per task.
BehaveRec study examines long-term behavioral history in recommendations
Researchers proposed BehaveRec, a framework that combines session, history, and community signals for sequential recommendations. The study used 137 million sequences from 12.6 million users to test when long-term history helps. Results show the benefit grows as users have more interactions, with a practical deployment threshold of about 20 interactions per user. The model is lightweight at 5.67 million parameters and trains in under two days on a single GPU.
Dubai AI Festival draws 20000 participants and global tech investment
The Dubai AI Festival brought together more than 20000 participants from 100 countries with 500 investors and 200 exhibitors. Over 100 Chinese technology companies and 23 Indian firms showcased their work at the event. The Dubai International Finance Center AI campus has attracted more than 500 AI companies and over AED 1 billion in investment since May 2024. Scale AI opened a new office at DIFC and signed a partnership with Abu Dhabi-based Presight to develop AI solutions for government and enterprise customers.
Sonar Hunter Agent finds logic bugs that SAST tools miss
Sonar released a new Hunter Agent that hunts for logic bugs in code that traditional SAST tools often miss. The agent maps a repository globally and runs specialized scans for broken access control, business logic, and authentication flaws. In a demo scanning a Python app in about one hour, it found 46 open issues including a high severity IDOR vulnerability. Scheduled scans can run daily weekly or monthly, and automatic remediation through a SonarQube agent becomes available this week.
Multi-agent system analyzes news framing across 17 outlets
Same Story Different Lens is a research prototype that compares how 17 news outlets cover the same events. It collects world and politics coverage then groups reports about the same event for structured analysis. Three AI agents handle article analysis, source support assessment, and cross-outlet briefing generation. The system produces analyses of sentiment political lean framing loaded language and possible omissions stored in a JSONL archive. The evaluation compares outlet-level lean estimates with reference ratings and includes model-assigned grounding scores.
Q2 Holdings stock surges on AI banking growth thesis
Q2 Holdings shares have gained 49.04% since a bullish investment thesis was published on June 5th. The growth was supported by higher revenue and profits along with raised full-year guidance and an expanded share repurchase authorization. The article compares Q2 Holdings against competitors PYPL and NCNO and explores whether AI-powered banking infrastructure can unlock its next growth phase.
California requires 15-minute handoff from AI customer service
California's AB 1609 takes effect January 1 2027 requiring businesses with over 500 million dollars in revenue to connect a customer with a person within 15 minutes. Fines can reach 5000 dollars per violation for companies that fail to comply. Salesforce Zendesk and Genesys are building the human handoff feature into their AI service products. New York City also proposed a rule requiring businesses to keep records of automated interactions for three years. Chatbots can still answer first but cannot be the only option and must disclose when they could pass for a human.
R2A-Sales benchmark reveals gap between sales agent rules and actions
Researchers created R2A-Sales, a benchmark that tests whether AI sales agents follow the rules they know when making real decisions. The study paired static policy questions with interactive sales scenarios using the same policy atoms across three Chinese-language domains. Static compliance scored 89.9 to 99.1 percent, but grounded action compliance fell to only 3.0 to 17.3 percent. The gap of 80.4 to 92.5 percentage points shows that measuring rule recall alone is not enough to ensure reliable sales agent performance.
Tesla renames FSD computers from AI to SI in owner manuals
Tesla has updated the owner's manuals for the Model 3, Model Y, and Cybertruck to replace the term FSD Computer with SI Computer. Under the new naming scheme, HW2.0, HW2.5, HW3, and AI4 become SI2.0, SI2.5, SI3, and SI4 respectively. The change follows Elon Musk's broader shift from the term AI to Super Intelligence. Although the rename has not been officially announced, the updated manuals suggest the new labels may appear on vehicle screens through a future software update.
Sources
- AI air cargo growth moderation expected
- AI investment will fuel air cargo demand, Boeing forecast says
- Artificial Analysis Launches Intelligence Index v5 With Majority Private Datasets, OpenAI Models Gain, Muse Wonder, Kimi Drop
- Announcing Artificial Analysis Intelligence Index v5
- BehaveRec: When Does Long-Term Behavioral History Help Sequential Recommendation? A Study with Session, History, and Community Signals
- Global AI leaders gather in Dubai as the emirate expands its ambitions
- Sonar Hunter Agent Hunts Logic Bugs SAST Misses
- Same Story, Different Lens: A Multi-Agent System for Cross-Outlet News Framing Analysis
- Q2 Holdings (QTWO) vs. PYPL and NCNO: Can AI-Powered Banking Infrastructure Unlock Its Next Growth Phase?
- California Puts a 15-Minute Clock on AI Customer Service Handoffs
- R2A-Sales: A Policy-Atom-Matched Benchmark for Rule-to-Action Gaps in Goal-Directed Sales Agents
- Goodbye AI4, Hello SI4: Tesla Renames Its FSD Computers
Comments
Please log in to post a comment.