Anthropic has raised concerns about Z.ai’s GLM-5.3, a Chinese frontier model released in August. In sandboxed tests the model found software vulnerabilities and built end-to-end exploits. On ExploitBench it succeeded in 50 of 410 attempts, close to Anthropic’s Claude Mythos Preview at 56. On an internal binary exploitation benchmark, it achieved full control-flow hijacks in 4 percent of tasks, something earlier models had not managed. Anthropic also showed researchers using the model to discover previously unknown browser flaws and chain them into working exploits with limited human time. The bigger issue, the company says, is that GLM-5.3 is open-weight. Users can download and modify the weights, making safeguards easier to weaken. Simple techniques raised engagement with harmful cyber requests from near zero to as high as 100 percent in tests. An “abliterated” version cut refusal rates dramatically while keeping much of its general capability.
AWS is expanding Aurora PostgreSQL so users can analyze historical data stored in Amazon S3 directly from the database. The feature embeds DuckDB inside Aurora, allowing queries that combine live operational tables with data held in Apache Iceberg and Parquet formats without first copying it into the database. Applications continue to use existing PostgreSQL interfaces. The change follows AWS’s recent acquisition of the company behind DuckDB. During queries Aurora reads only the necessary data and columns from S3 and can cache frequently used data. Developers can see metrics such as rows scanned, data read from S3, and cache hits. AWS illustrated the capability with a financial example that joined seven days of recent transactions in Aurora with five years of history in a Parquet file on S3. External Iceberg catalogs can be linked via AWS Glue. There is no separate feature fee, though compute and S3 read costs may rise.
Salesforce has signed a definitive agreement to acquire Listen Labs, an AI-powered customer research and human simulation platform. Financial terms were not disclosed, though reports put the value near $2 billion. The deal is expected to close in Salesforce’s fiscal fourth quarter of 2027, subject to regulatory approval. Listen Labs uses AI to design studies, recruit participants, conduct interviews, and analyze responses. It draws on a network of more than 50 million research participants and can run interviews in more than 120 languages. The platform also builds digital twins—AI simulations grounded in real customer behavior—so teams can test how people might respond to products or messages before further human testing. Listen Labs will join Salesforce AI Labs. Salesforce says the technology will complement Marketing Cloud and Service Cloud and supply richer context to its AI agents, turning qualitative feedback into insights and actions at scale.
Researchers at the Singapore University of Technology and Design and MIT have built ALBATROSS, a bio-inspired robot that drops from the air, lands softly on water and converts into an autonomous sailboat. Short for Airborne Lander with Buoyant Auto-Rotating Sailing Sensor, the roughly 1-kilogram platform uses two rigid wingsails that spin passively like a maple seed during a 150-meter descent. The autorotation slows the fall to about 7 metres per second and cuts impact energy eighteen-fold, allowing a controlled water landing without heavy motors or a parachute. On impact the craft passively self-rights. The same wings then act as wind-driven sails. A fish-fin-style tail rudder provides steering and, in low wind, oscillating propulsion up to 0.4 metres per second. The design uses only three actuators and three main sensors and achieves 91 percent effective mass usage across air and water. In trials at two Singapore reservoirs the robot sailed autonomously.
BMW plans to cut around 20 percent of its senior management positions by mid-2027. The German automaker will streamline divisions and management structures under an agreed buyout program. It currently has about 65 senior vice presidents reporting to the board and roughly 400 senior management roles beneath them. The changes could affect around 100 high-level positions, most of them in Munich. BMW says greater use of artificial intelligence across operations will support leaner structures, faster decision-making, and lower costs. Chief Financial Officer Walter Mertl described consistent use of AI agents as a game-changer for more agile development and efficient operations. The move follows a wider voluntary departure program announced earlier that could eliminate about 8,000 white-collar jobs, or roughly 5 percent of the global workforce. BMW is under pressure from weaker demand in China and has warned its automotive profit margin could fall as low as 1 percent this year.
Google has released Gemini 4 Argon, which it calls its most powerful AI model yet. The model is designed for complex, long-horizon work, including coding, research, writing, and enterprise knowledge tasks in areas such as legal and finance. Cybersecurity is a particular focus. Argon was trained for defensive cyber work and can autonomously find, validate, and patch critical software vulnerabilities. It is rolling out first to a select group of cyber partners through Google’s Fairwind Program. Google staff are already using the model internally for debugging, codebase migrations, and other daily engineering tasks. The company also highlights Argon’s ability to analyze long videos and charts. Output capacity has been raised to an industry-leading 1 million tokens, up from 64,000 on earlier Gemini models. Google says Argon scores higher than OpenAI’s GPT-6 Astra and Anthropic’s Fable and Opus models on multiple benchmarks and leads the Vals Index.
Indian enterprises are moving agentic AI from experimentation into production and must now prove measurable business value. According to IDC, India’s AI spending is projected to reach USD 6 billion, growing at a CAGR of 33.7% through 2027. Leaders need frameworks that go beyond cost reduction or productivity claims. Feedback comes in two forms: explicit signals such as thumbs-up or thumbs-down and implicit signals drawn from outcomes, for example, whether a resolved incident stays closed or a workflow finishes without further intervention. Implicit feedback often better reflects real business value. Metrics should match the agent’s role and be refined over time. Deployments typically start under close human supervision. Low-risk tasks can later run with greater autonomy, while high-impact actions involving finance, critical infrastructure, or sensitive data should still require approval. Continuous evaluation and clear governance will decide which programs scale with confidence.
Google is paying about 100 publishers in its AI contribution pilot for content used in AI Overviews, AI Mode, and Gemini. According to reporting based on The Information, payments for several small- to midsize-sized sites amount to less than one-tenth of 1% of their advertising revenue. Payouts vary widely. One publisher that joined a few months ago has earned about $50,000 to $60,000 so far. Another early participant is earning more than $1 million a year. Smaller sites have received less than $1,000 over several months. Payments are said to reflect how much a source contributes to an AI-generated answer, but some participants say they do not know how Google calculates the figures and that amounts change month to month without explanation. Publishers manage participation and track earnings in Google Search Console. Content qualifies only when it significantly influences the generation of a response; later fact-checking or linking does not count.
Google is testing an AI-powered shopping experience in India that lets select users buy products from Walmart-owned Flipkart directly through Gemini and AI Mode. Users in the pilot see a Buy button on selected Flipkart product listings. Tapping it opens a Flipkart-branded checkout flow so shoppers can complete the purchase without leaving the AI interface or searching separately on Flipkart. The early test is limited to some users and a small set of products, including smartphones, electronics, and mobile accessories. Other users still see Flipkart listings without the direct purchase option. Google reportedly plans a broader rollout later in October, ahead of India’s festive shopping season. The approach differs from Google’s Universal Commerce Protocol work, which used a Google-hosted checkout; the technology behind the Flipkart integration has not been disclosed.
Only 51% of employees have access to the learning and development resources they need, down from 59% a year earlier, according to PwC’s 2026 Global Workforce Hopes and Fears Survey. The study covered 49,364 workers across 48 countries and regions. While 61% of employees believe they can develop new skills, about half say they lack the resources to do so. The gap is sharpest among “engine room” workers, who make up 56% of the workforce and have less scarce skills and lower AI maturity; only two in five in this group report adequate learning support. AI use is rising: 64% of workers used AI at work in the past year, up 10 percentage points, and daily generative AI use climbed from 14% to 22%. Daily AI users report higher confidence in job security, promotions, and learning ability. PwC also identified four workforce groups, including front-runners (14%) with scarce skills and strong AI capabilities, nearly 29% of whom are likely to change employers within a year.
Anthropic has raised concerns about Z.ai’s GLM-5.3, a Chinese frontier model released in August. In sandboxed tests the model found software vulnerabilities and built end-to-end exploits. On ExploitBench it succeeded in 50 of 410 attempts, close to Anthropic’s Claude Mythos Preview at 56. On an internal binary exploitation benchmark, it achieved full control-flow hijacks in 4 percent of tasks, something earlier models had not managed. Anthropic also showed researchers using the model to discover previously unknown browser flaws and chain them into working exploits with limited human time. The bigger issue, the company says, is that GLM-5.3 is open-weight. Users can download and modify the weights, making safeguards easier to weaken. Simple techniques raised engagement with harmful cyber requests from near zero to as high as 100 percent in tests. An “abliterated” version cut refusal rates dramatically while keeping much of its general capability.
AWS is expanding Aurora PostgreSQL so users can analyze historical data stored in Amazon S3 directly from the database. The feature embeds DuckDB inside Aurora, allowing queries that combine live operational tables with data held in Apache Iceberg and Parquet formats without first copying it into the database. Applications continue to use existing PostgreSQL interfaces. The change follows AWS’s recent acquisition of the company behind DuckDB. During queries Aurora reads only the necessary data and columns from S3 and can cache frequently used data. Developers can see metrics such as rows scanned, data read from S3, and cache hits. AWS illustrated the capability with a financial example that joined seven days of recent transactions in Aurora with five years of history in a Parquet file on S3. External Iceberg catalogs can be linked via AWS Glue. There is no separate feature fee, though compute and S3 read costs may rise.
Salesforce has signed a definitive agreement to acquire Listen Labs, an AI-powered customer research and human simulation platform. Financial terms were not disclosed, though reports put the value near $2 billion. The deal is expected to close in Salesforce’s fiscal fourth quarter of 2027, subject to regulatory approval. Listen Labs uses AI to design studies, recruit participants, conduct interviews, and analyze responses. It draws on a network of more than 50 million research participants and can run interviews in more than 120 languages. The platform also builds digital twins—AI simulations grounded in real customer behavior—so teams can test how people might respond to products or messages before further human testing. Listen Labs will join Salesforce AI Labs. Salesforce says the technology will complement Marketing Cloud and Service Cloud and supply richer context to its AI agents, turning qualitative feedback into insights and actions at scale.
Researchers at the Singapore University of Technology and Design and MIT have built ALBATROSS, a bio-inspired robot that drops from the air, lands softly on water and converts into an autonomous sailboat. Short for Airborne Lander with Buoyant Auto-Rotating Sailing Sensor, the roughly 1-kilogram platform uses two rigid wingsails that spin passively like a maple seed during a 150-meter descent. The autorotation slows the fall to about 7 metres per second and cuts impact energy eighteen-fold, allowing a controlled water landing without heavy motors or a parachute. On impact the craft passively self-rights. The same wings then act as wind-driven sails. A fish-fin-style tail rudder provides steering and, in low wind, oscillating propulsion up to 0.4 metres per second. The design uses only three actuators and three main sensors and achieves 91 percent effective mass usage across air and water. In trials at two Singapore reservoirs the robot sailed autonomously.
BMW plans to cut around 20 percent of its senior management positions by mid-2027. The German automaker will streamline divisions and management structures under an agreed buyout program. It currently has about 65 senior vice presidents reporting to the board and roughly 400 senior management roles beneath them. The changes could affect around 100 high-level positions, most of them in Munich. BMW says greater use of artificial intelligence across operations will support leaner structures, faster decision-making, and lower costs. Chief Financial Officer Walter Mertl described consistent use of AI agents as a game-changer for more agile development and efficient operations. The move follows a wider voluntary departure program announced earlier that could eliminate about 8,000 white-collar jobs, or roughly 5 percent of the global workforce. BMW is under pressure from weaker demand in China and has warned its automotive profit margin could fall as low as 1 percent this year.
Google has released Gemini 4 Argon, which it calls its most powerful AI model yet. The model is designed for complex, long-horizon work, including coding, research, writing, and enterprise knowledge tasks in areas such as legal and finance. Cybersecurity is a particular focus. Argon was trained for defensive cyber work and can autonomously find, validate, and patch critical software vulnerabilities. It is rolling out first to a select group of cyber partners through Google’s Fairwind Program. Google staff are already using the model internally for debugging, codebase migrations, and other daily engineering tasks. The company also highlights Argon’s ability to analyze long videos and charts. Output capacity has been raised to an industry-leading 1 million tokens, up from 64,000 on earlier Gemini models. Google says Argon scores higher than OpenAI’s GPT-6 Astra and Anthropic’s Fable and Opus models on multiple benchmarks and leads the Vals Index.
Indian enterprises are moving agentic AI from experimentation into production and must now prove measurable business value. According to IDC, India’s AI spending is projected to reach USD 6 billion, growing at a CAGR of 33.7% through 2027. Leaders need frameworks that go beyond cost reduction or productivity claims. Feedback comes in two forms: explicit signals such as thumbs-up or thumbs-down and implicit signals drawn from outcomes, for example, whether a resolved incident stays closed or a workflow finishes without further intervention. Implicit feedback often better reflects real business value. Metrics should match the agent’s role and be refined over time. Deployments typically start under close human supervision. Low-risk tasks can later run with greater autonomy, while high-impact actions involving finance, critical infrastructure, or sensitive data should still require approval. Continuous evaluation and clear governance will decide which programs scale with confidence.
Google is paying about 100 publishers in its AI contribution pilot for content used in AI Overviews, AI Mode, and Gemini. According to reporting based on The Information, payments for several small- to midsize-sized sites amount to less than one-tenth of 1% of their advertising revenue. Payouts vary widely. One publisher that joined a few months ago has earned about $50,000 to $60,000 so far. Another early participant is earning more than $1 million a year. Smaller sites have received less than $1,000 over several months. Payments are said to reflect how much a source contributes to an AI-generated answer, but some participants say they do not know how Google calculates the figures and that amounts change month to month without explanation. Publishers manage participation and track earnings in Google Search Console. Content qualifies only when it significantly influences the generation of a response; later fact-checking or linking does not count.
Google is testing an AI-powered shopping experience in India that lets select users buy products from Walmart-owned Flipkart directly through Gemini and AI Mode. Users in the pilot see a Buy button on selected Flipkart product listings. Tapping it opens a Flipkart-branded checkout flow so shoppers can complete the purchase without leaving the AI interface or searching separately on Flipkart. The early test is limited to some users and a small set of products, including smartphones, electronics, and mobile accessories. Other users still see Flipkart listings without the direct purchase option. Google reportedly plans a broader rollout later in October, ahead of India’s festive shopping season. The approach differs from Google’s Universal Commerce Protocol work, which used a Google-hosted checkout; the technology behind the Flipkart integration has not been disclosed.
Only 51% of employees have access to the learning and development resources they need, down from 59% a year earlier, according to PwC’s 2026 Global Workforce Hopes and Fears Survey. The study covered 49,364 workers across 48 countries and regions. While 61% of employees believe they can develop new skills, about half say they lack the resources to do so. The gap is sharpest among “engine room” workers, who make up 56% of the workforce and have less scarce skills and lower AI maturity; only two in five in this group report adequate learning support. AI use is rising: 64% of workers used AI at work in the past year, up 10 percentage points, and daily generative AI use climbed from 14% to 22%. Daily AI users report higher confidence in job security, promotions, and learning ability. PwC also identified four workforce groups, including front-runners (14%) with scarce skills and strong AI capabilities, nearly 29% of whom are likely to change employers within a year.