Search “Anthropic”

Anthropic releases report disclosing changes in the company's workforce structure

Anthropic released an internal research report analyzing AI's impact on the company's hiring and work organization, based on data including employment, AI usage, output metrics, employee surveys, calendars and recruiting records, as well as interviews with 17 employees. The report says the share of junior employees in technical departments fell from 21% in early 2025 to 14%, and the share of junior technical roles in hiring dropped from 27% to 9%; by August, interactive AI agents' working hours were six times those of employees, and per-engineer code commits reached five times the level of six months earlier. The report also notes that human attention has become the bottleneck for agent output, that employees are shifting toward meta-work of building monitoring and orchestration agents, and that hiring now places more emphasis on autonomy and judgment. Subsequent reporting says that 80% of the time spent in Claude sessions is agents waiting for human input, that in March agents' total working hours exceeded humans' for the first time, and that the share of interactive sessions employees use to manage agents rose from 10% to 22%.

Anthropic in talks with law firms on AI agent negotiation standards

Anthropic is working with law firms to develop fair negotiation standards for AI agents that bargain on users' behalf, handle complaints, or improve contract offers. According to the Financial Times, the company wants to embed lawyers' negotiation ethics into Claude agents so they can negotiate with other people's agents, resolve disputes, and discuss salaries. The report cited experimental data showing that in Project Deal, the Claude agents of 69 employees completed 186 transactions without human confirmation, totaling slightly more than $4,000; Opus agents sold the same goods for $2.68 more and bought at $2.45 less than Haiku, yet Haiku users rated fairness almost the same as Opus users, at 4.06 and 4.05 respectively on a 7-point scale. Project Swap showed that instructions had far less impact than the model itself: using aggressive instructions raised the score by only 0.02, while switching from Haiku to Opus raised it by 0.12. The report also said Anthropic is recruiting Conceptual Reasoning Fellows with legal backgrounds to spend four months testing Claude on judgment-intensive questions in law, philosophy, cognitive science, economics, and political science. The standards are still under discussion.

Anthropic and law firms discuss rules for AI agent negotiations

Anthropic is working with law firms to develop fair negotiation standards for AI agents that bargain on users' behalf, handle complaints, or improve contract offers. According to the Financial Times, the company wants to embed lawyers' negotiation ethics into Claude agents so they can negotiate with other people's agents, resolve disputes, and discuss salaries. The report cited experimental data showing that in Project Deal, the Claude agents of 69 employees completed 186 transactions without human confirmation, totaling slightly more than $4,000; Opus agents sold the same goods for $2.68 more and bought at $2.45 less than Haiku, yet Haiku users rated fairness almost the same as Opus users, at 4.06 and 4.05 respectively on a 7-point scale. Project Swap showed that instructions had far less impact than the model itself: using aggressive instructions raised the score by only 0.02, while switching from Haiku to Opus raised it by 0.12. The report also said Anthropic is recruiting Conceptual Reasoning Fellows with legal backgrounds to spend four months testing Claude on judgment-intensive questions in law, philosophy, cognitive science, economics, and political science. The standards are still under discussion.

View story:Anthropic in talks with law firms on AI agent negotiation standards

English translation is being prepared.

English translation is being prepared.

View story:English translation is being prepared.

Anthropic red team says it created a self-propagating AI worm

A report released this week by Anthropic's offensive research team claims to have created a malicious AI program that can spread autonomously among internet-connected smart TVs, splitting itself across 15 TVs and borrowing a little computing power from each to keep running. According to The Guardian's interpretation, the researchers believe a successful AI worm needs stronger capabilities on lower-powered devices, but that current trends suggest this could be feasible by the end of the year; a worm in an earlier paper from the University of Toronto infected a full-sized computer and spread faster. Security experts are divided: some stress that IoT devices such as smart TVs are slow to receive patches and their risks are hard to fix, while others argue the experiment moves slowly and its actions are predictable and interceptable, making it stunt hacking. At present, this claim appears only in the above report, with no further developments.

View story:Anthropic red team says it created a self-propagating AI worm

English translation is being prepared.

English translation is being prepared.

View story:English translation is being prepared.

Anthropic announces research on Mythos 5 autonomously designing quantum circuits

On October 17, 2026, Anthropic published a study: when given only a pure natural-language specification and an automated evaluator, Mythos 5 essentially without human intervention repeatedly submitted and improved quantum circuit designs in a closed loop, finding circuits that use fewer qubits and gates than the best previously published solutions, with some improvements reaching an order of magnitude. The elliptic-curve modular inversion design came from a single 53-hour continuous run, in which the model derived a reversible divsteps implementation that requires no storage of branch information and formally verified it with Lean; another effort reduced the gate count of the Reed-Solomon decoding step, the most expensive step in DQI, to about 1/12 of a construction previously made by Google. Anthropic also noted that the model was initially conservative and tended to follow known methods, which needed to be mitigated by requiring parallel exploration of multiple approaches and occasionally prompting it to broaden its search. Subsequent reports said the research came from the STEM Fellows program and gave differing formulations, such as a 17-fold reduction in gate count for the divsteps task.

Anthropic releases report disclosing changes in the company's employment structure

Anthropic released an internal research report analyzing AI's impact on the company's hiring and work organization, based on data including employment, AI usage, output metrics, employee surveys, calendars and recruiting records, as well as interviews with 17 employees. The report says the share of junior employees in technical departments fell from 21% in early 2025 to 14%, and the share of junior technical roles in hiring dropped from 27% to 9%; by August, interactive AI agents' working hours were six times those of employees, and per-engineer code commits reached five times the level of six months earlier. The report also notes that human attention has become the bottleneck for agent output, that employees are shifting toward meta-work of building monitoring and orchestration agents, and that hiring now places more emphasis on autonomy and judgment. Subsequent reporting says that 80% of the time spent in Claude sessions is agents waiting for human input, that in March agents' total working hours exceeded humans' for the first time, and that the share of interactive sessions employees use to manage agents rose from 10% to 22%.

View story:Anthropic releases report disclosing changes in the company's workforce structure

Anthropic engineer shares multi-agent orchestration strategies

On October 20, 2026, an Anthropic engineer shared advanced usage of Claude Code in a 30-minute workshop: running more than 100 agents simultaneously within one integrated loop, with 1 main agent overseeing while sub-agents act as project managers to break down tasks, distribute them, and cross-verify results. Based on this, the related report advises developers to first write task decomposition and acceptance rules in the main window, then spawn sub-agents to execute submodules concurrently; it also cautions that when a hundred agents run at once, if termination conditions and permission boundaries are not hard-coded, it is easy to consume a large amount of Tokens in an infinite loop, and beginners should start small with one main and three sub-agents. Currently, this sharing has been reported by a single source, and there are no subsequent factual developments.

Anthropic red team says it created a self-propagating AI worm

A report released this week by Anthropic's offensive research team claims to have created a malicious AI program that can spread autonomously among internet-connected smart TVs, splitting itself across 15 TVs and borrowing a little computing power from each to keep running. According to The Guardian's interpretation, the researchers believe a successful AI worm needs stronger capabilities on lower-powered devices, but that current trends suggest this could be feasible by the end of the year; a worm in an earlier paper from the University of Toronto infected a full-sized computer and spread faster. Security experts are divided: some stress that IoT devices such as smart TVs are slow to receive patches and their risks are hard to fix, while others argue the experiment moves slowly and its actions are predictable and interceptable, making it stunt hacking. At present, this claim appears only in the above report, with no further developments.

Anthropic co-founder proposes AI may drive US GDP to $44 trillion by 2030

Anthropic co-founder Jack Clark proposed that, under certain conditions, AI could drive US GDP to $44 trillion by 2030. A report on October 18, 2026 stated that Clark, citing the company's internal model forecasts, said AI could boost US economic output by nearly 50%, bringing the potential scale of US GDP to $44 trillion; he also warned that growing evidence points to a "technological singularity," that AI development will outpace human control, and that the window to build safety safeguards is only "a few short years." A report on October 23 further stated that Clark, during the IMF and World Bank annual meetings, cited internal model projections in which Anthropic's economic research set out three scenarios: a moderate scenario of $34.1 trillion, a significant scenario of $36.3 trillion, and a most aggressive scenario of $44.4 trillion, with the extreme scenario representing an increase of about 32.4% relative to a baseline without AI acceleration. The report noted that these results are not deterministic forecasts, and whether they materialize depends on technological progress, the pace of corporate adoption, and the actual effects of productivity gains.

View story:Anthropic co-founder suggests AI could drive US GDP to $44 trillion by 2030

Anthropic publishes "The AI Expectations Gap" survey

On October 16, 2026, Anthropic published the "The AI Expectations Gap" survey, comparing the expectations of U.S. adults and Anthropic employees regarding AI. The study is based on a survey conducted in August 2026. Official evidence shows that 91% of Anthropic employees believe AI will be able to make Nobel Prize-level discoveries by 2030, compared with 39% of U.S. adults; employees expect on average that AI will be able to complete 75% of their current work by 2030, compared with 28% for employed Americans. According to the study's economic model, the median U.S. adult's responses correspond to 5.4% GDP growth in 2030, while the median employee's correspond to 17.1%. The two groups differ markedly on the share of their own work that will be replaced by AI, but both believe that the more replacement there is, the higher the unemployment rate will be. Subsequent reporting said that the two groups largely agree in their judgments of AI's current capabilities, with the divergence concentrated on the pace of technological progress and the scope of corporate adoption in the coming years; 95% of Anthropic employees still believe the company's total headcount will increase over the next 12 months, with a median expected growth of 66%. The survey has now been publicly released, with no further developments.

Anthropic uses gzip to reduce Claude Code latency

Anthropic improved the latency performance of Claude Code. According to an Analytics India Magazine report on October 19, 2026, Anthropic said on X that Claude Code now compresses with gzip before sending each conversation turn, so content reaches the model faster; cloud sessions also start as soon as the user begins typing. The report said that compared with August, local sessions are 3 times faster and cloud sessions 2.8 times faster, and the API, chat and Cowork platforms are 1.5 times faster. The report also explained that these figures are measured by p90 first-token latency and exclude model processing time; the Claude API figures use p50 first-token time including model processing time. Anthropic has publicly described this improvement, and there are no further developments so far.

Anthropic publishes "The AI Expectations Gap" survey

On October 16, 2026, Anthropic published the "The AI Expectations Gap" survey, comparing the expectations of U.S. adults and Anthropic employees regarding AI. The study is based on a survey conducted in August 2026. Official evidence shows that 91% of Anthropic employees believe AI will be able to make Nobel Prize-level discoveries by 2030, compared with 39% of U.S. adults; employees expect on average that AI will be able to complete 75% of their current work by 2030, compared with 28% for employed Americans. According to the study's economic model, the median U.S. adult's responses correspond to 5.4% GDP growth in 2030, while the median employee's correspond to 17.1%. The two groups differ markedly on the share of their own work that will be replaced by AI, but both believe that the more replacement there is, the higher the unemployment rate will be. Subsequent reporting said that the two groups largely agree in their judgments of AI's current capabilities, with the divergence concentrated on the pace of technological progress and the scope of corporate adoption in the coming years; 95% of Anthropic employees still believe the company's total headcount will increase over the next 12 months, with a median expected growth of 66%. The survey has now been publicly released, with no further developments.

View story:Anthropic publishes "The AI Expectations Gap" survey

Anthropic engineer shares multi-agent orchestration approach at workshop

On October 20, 2026, an Anthropic engineer shared advanced usage of Claude Code in a 30-minute workshop: running more than 100 agents simultaneously within one integrated loop, with 1 main agent overseeing while sub-agents act as project managers to break down tasks, distribute them, and cross-verify results. Based on this, the related report advises developers to first write task decomposition and acceptance rules in the main window, then spawn sub-agents to execute submodules concurrently; it also cautions that when a hundred agents run at once, if termination conditions and permission boundaries are not hard-coded, it is easy to consume a large amount of Tokens in an infinite loop, and beginners should start small with one main and three sub-agents. Currently, this sharing has been reported by a single source, and there are no subsequent factual developments.

View story:Anthropic engineer shares multi-agent orchestration strategies

Anthropic uses gzip to reduce Claude Code latency

Anthropic improved the latency performance of Claude Code. According to an Analytics India Magazine report on October 19, 2026, Anthropic said on X that Claude Code now compresses with gzip before sending each conversation turn, so content reaches the model faster; cloud sessions also start as soon as the user begins typing. The report said that compared with August, local sessions are 3 times faster and cloud sessions 2.8 times faster, and the API, chat and Cowork platforms are 1.5 times faster. The report also explained that these figures are measured by p90 first-token latency and exclude model processing time; the Claude API figures use p50 first-token time including model processing time. Anthropic has publicly described this improvement, and there are no further developments so far.

View story:Anthropic uses gzip to reduce Claude Code latency

English translation is being prepared.

English translation is being prepared.

View story:English translation is being prepared.