AI news, week of October 5, 2026

Machine summary DCSA has disavowed generative AI, large language models, facial action coding, and vocal stress analysis in the Modernizing Polygraph program, but left deceptive-speech language models unaddressed. The move follows scientific and civil-liberty criticism of AI-driven trustworthiness scoring.

1. The agency also said the research does not involve "facial action coding [of emotions], micro-expression analysis or vocal analyses."
2. As of July, declassified presentation slides and prior DoD statements showed AI tools being prepared to analyze text, voices, and faces to estimate trustworthiness.
3. DCSA said the DoD will "continue to monitor emerging scientific capabilities to evaluate their potential utility down the road."
4. DCSA did not address other AI models shown in the slides that researchers were developing to scan spoken and written language for "deceptive speech" patterns.
5. National security lawyers, scientists, and AI auditors warned that AI-aided deception detection lacks scientific backing, would compound errors, and may violate civil liberties.

Reported by Defense News and militarytimes.com.

• Large Language Models
• Micro-expression analysis
• AI polygraph analysis
• Vocal stress analysis
• Modernizing Polygraph
• National security lawyers, scientists, and AI auditors

Reports ran from 21:44 UTC on Oct 5 to 23:38 UTC on Oct 6.

The outlets that reported this event are listed with links to their reports; no summary of ours passed our automated checks for it, which says nothing about the event (how they work).

Reported by Bloomberg and The Star.

• GPU

Machine summary Meta announced a range of advertising updates at its Advertising Week presentation in New York, including an improved AI-powered business assistant, expanded creative tools for brands and new shopping experiences, according to Social Media Today. The company said it is testing importing tools and scheduled tasks, and is expanding availability of Ads Creative Studio and its Customer Lifecycle Strategy.

1. Meta announced the updates at its Advertising Week presentation in New York, according to Social Media Today.
2. Meta said it is adding new creative insights and giving its AI-powered business assistant more memory and context, including what advertisers tried last quarter, seasonality and stated goals.
3. Meta said it is testing importing tools that will let brands upload strategy and brand voice guidelines to refine outputs.
4. Meta stated that it is trying out scheduled tasks, including alerts and recurring performance reports.
5. Meta stated that Advantage+ creative now has the ability to create video from static assets that already exist, and Meta said that feature is generally available.

Reported by Social Media Today and Marketing Dive.

• Meta
• Ads Creative Studio
• Advantage+ creative
• Customer Lifecycle Strategy

Reports ran between 03:38 and 10:14 UTC on Oct 8.

The outlets that reported this event are listed with links to their reports; no summary of ours passed our automated checks for it, which says nothing about the event (how they work).

Reported by The Hindu, timeslive.co.za, Seeking Alpha and thenextweb.com.

• CrowdStrike

Machine summary TechCrunch reports that Amazon announced three new Alexa tablets — the Alexa Tablet 12 Pro, Alexa Tablet 11 and Alexa Tablet 8 — which run Android and offer access to the Google Play Store. An Amazon spokesperson told TechCrunch the devices are a new product line with different hardware and software rather than a rebranded Fire tablet. TechCrunch reported the tablets start at $229.99, $329.99 and $499.99 depending on the model, with availability in the U.S., Mexico and Canada starting Oct. 14, 2026.

Reported by techcrunch.com and Seeking Alpha.

• Amazon
• Google
• Alexa+
• Google Play Store
• Alexa

Reports ran between 20:59 and 21:47 UTC on Oct 10.

The outlets that reported this event are listed with links to their reports; no summary of ours passed our automated checks for it, which says nothing about the event (how they work).

Reported by CNBC and techcrunch.com.

• Microsoft
• Satya Nadella
• AI Models

Machine summary Korea's AI-led stock surge has reversed sharply as memory-chip concentration, foreign outflows, and peak-cycle fears mount. The Kospi's 22% second-half slump and $131 billion in foreign withdrawals signal cooling confidence in the AI memory supply chain that powers accelerators.

1. The Kospi was the best-performing major equities benchmark in the first half of 2026, then lost 22% in the second half to become the worst.
2. Turnover in South Korea's $4.3 trillion stock market has collapsed 70% from its late-May peak.
3. Foreign investors pulled $131 billion from Korean stocks this year, the most among major Asian markets.
4. Samsung Electronics and SK Hynix account for more than half of the Kospi's index weighting.
5. Samsung shares fell despite reporting a nearly nine-fold surge in quarterly operating profit.

Reported by Bloomberg and economictimes.indiatimes.com.

• SK Hynix
• South Korea
• Samsung Electronics Co.
• Rayliant Global Advisors

Machine summary Sakana AI's MASS framework demonstrates an AI system using multi-agent collective intelligence to recursively improve itself without relying on human evaluators.

1. MASS (Multi-Agent Self-Supervision) optimizes multi-agent workflows and distills collective intelligence into a self-supervision signal for recursive self-improvement.
2. The framework positions the same AI model as optimizee, optimizer, and evaluator, an approach the authors call homogeneous recursive self-improvement.
3. On research benchmarks, MASS improves score per output token over the base model, though the available summary does not disclose exact percentage gains.
4. The work was done during an internship at Sakana AI, with Hyunin Lee credited as a contributing researcher.
5. The paper was published on Sakana AI's research blog on October 11, 2026, with an estimated reading time of 15 minutes.

Reported by pub.sakana.ai.

• Recursive self-improvement
• Sakana AI
• MASS
• Collective intelligence

Machine summary He warns that without regulation comparable to aviation or medicine, AI will amplify harmful capabilities from bioweapons to cyberattacks.

1. Gates said "we're not doing enough" to keep AI out of the wrong hands during a CNN interview with Jake Tapper on October 11, 2026.
2. He argued humanity hasn't found a "middle ground" for safeguarding AI despite comparing needed regulation to planes, cars, and medicines.
3. Gates said the risk of a bioterrorist attack is now 50 times the risk of a natural pandemic, according to Tapper's recap of an earlier warning.
4. Gates predicted cyber attacks will be "the front that we'll probably see most dramatically in the next year."
5. He warned bad actors using the latest models without safeguards will produce "a lot of negative surprises" over the next year.

Reported by dailymail.com.

• Sam Altman
• Dario Amodei
• CNN
• Bill Gates